Consumer Terms of Service
Governs Claude Free/Pro/Max; permits model-improvement use unless user opts out, with feedback/safety exceptions and organization-account controls.
“train our models”https://www.anthropic.com/legal/consumer-terms
00.1
WBS/PBS 00.1
Executive answer
Privacy is a stack of contracts, retention rules, safety systems, employer controls and hosting boundaries—not one “private mode”.
Anthropic privacy atlas · v01 · checked 2026-10-03
“Perfect privacy” is not a Claude plan.
You can get no model training, zero prompt/response storage at rest for eligible API paths, customer-controlled encryption keys, access logging, private networking and cloud-provider isolation. You cannot make a remote AI service process sensitive data while simultaneously guaranteeing that no external system processes it.
Consumer privacy ceilingOpt-out + IncognitoStill safety/legal retention paths.
Direct Anthropic ceilingAPI ZDR + controlsNot the normal Enterprise Chat UI.
Anthropic-content isolation ceilingAWS BedrockAWS says content is not shared with model providers.
Core evidence SRC-ANT-003 SRC-ANT-004 SRC-ANT-010 SRC-AWS-001 SRC-ANT-023
YOUR
DATA
DATA
Anthropicmodel / safety / product
Cloud hostAWS · GCP · Azure
Your employerowners · compliance
Connectorsthird-party terms
Metadataaccount · usage · logs
Law / safetyexception paths
DIA-00-001 · Privacy is a multi-actor system, not one switchdeterministic inline SVG
Critical distinction: “not used for training” ≠ “not stored”; “ZDR” ≠ “no automated processing”; “private from Anthropic” ≠ “private from your employer or cloud provider.”
Source packet for this tab 6 sources
Commercial Terms of Service
Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract.
“may not train models”https://www.anthropic.com/legal/commercial-terms
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Anthropic Trust Center FAQ
States Claude for Work is multi-tenant and Anthropic does not offer single-tenancy; customer data is logically separated.
“multi-tenant architecture”https://trust.anthropic.com/faq
Amazon Bedrock FAQs — Security
AWS states prompts/outputs are not shared with model providers, not used for base-model training, and can use PrivateLink.
“not shared with any model providers”https://aws.amazon.com/bedrock/faqs/
Who owns and manages the data of my team?
Explains employer/organization control: Primary Owner can manage and export member data including chats/files/usage.
“Primary Owner manages”https://support.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team
10.1
WBS/PBS 10.1
Consumer: Free, Pro, Max
Paying more buys capacity and features; it does not buy a different consumer privacy regime.
Free/Pro/Maxconsumer
Incognitono training
Teamcommercial
Enterprise UIcontrols
API standardstateless
API ZDRno content at rest
Bedrock noneAWS boundary
Analytical privacy-from-Anthropic index, not an Anthropic rating. It measures documented content exposure/storage only; higher does not mean “private from your employer/cloud provider.”
What improves with price
Usage capacity: Free is limited; Pro is standard; Max is 5× or 20× Pro per session. Max resets the session limit every five hours and also has weekly/further discretionary caps.
What does not materially improve
Free, Pro and Max remain consumer products. Current pricing marks model training as opt-out. If model improvement is enabled, de-identified training-pipeline retention can be up to five years; safety flags can produce longer retention regardless.
10 rows
| Surface | Price / seat | Usage cap | Training | Normal content retention | Training retention | Safety exception | ZDR? | Org/admin visibility | Regime |
|---|---|---|---|---|---|---|---|---|---|
| Free | $0 | Limited | Opt-out | Until delete; backend ≤30d after delete | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Pro | $20/mo or $200/y | Standard | Opt-out | Same consumer policy | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Max 5x | $100/mo | 5× Pro per session; 5h reset + weekly | Opt-out | Same consumer policy | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Max 20x | $200/mo | 20× Pro per session; 5h reset + weekly | Opt-out | Same consumer policy | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Team Standard | $20 annual / $25 monthly per seat | More than Pro; weekly | No training by default | Saved until delete; backend ≤30d after delete | No by default | ≤2y flagged; scores ≤7y | No | Primary Owner can export | Commercial |
| Team Premium | $100 annual / $125 monthly per seat | 5× Team Standard | No training by default | Same Team policy | No by default | ≤2y flagged; scores ≤7y | No | Primary Owner can export | Commercial |
| Enterprise UI | $20/seat/mo annual + API-rate usage | Usage-based: no plan/seat cap; spend caps | No training by default | Default indefinite; custom minimum 30d | No by default | Safety/legal exceptions | No (UI not ZDR) | Compliance API / exports | Commercial |
| Claude API — standard | API rates | Rate-limit tier / spend based | No training by default | Stateless conversation content not retained by default; outer policy ≤30d when retained | No by default | Safety/legal + feature state | No | Org API/admin logs | Commercial |
| Claude API — ZDR | API rates | Rate-limit tier / spend based | No training by default | No prompt/response at rest after response | No | Classifier/legal/safety residue | Yes, eligible endpoints | Org logs; Compliance data separate | Commercial |
| AWS Bedrock — retention none | AWS + model pricing | Quota based | No training | Zero retention where model/feature permits | No | Covered-model review rules can override | AWS retention mode | AWS org/IAM logs | Partner cloud |
Best consumer configuration: turn model improvement off, use Incognito for sensitive one-off chats, do not submit thumbs feedback containing sensitive content, and avoid copying secrets into connectors. This is still not contractual ZDR.
Source packet for this tab 7 sources
Plans & Pricing
Current consumer, Team and Enterprise prices, relative usage multipliers, model-training defaults and enterprise controls.
“Opt-out”https://claude.com/pricing
Choose a Claude plan
Consumer plan price and usage-capacity comparison.
“Limited”https://support.claude.com/en/articles/11049762-choose-a-claude-plan
Consumer Terms of Service
Governs Claude Free/Pro/Max; permits model-improvement use unless user opts out, with feedback/safety exceptions and organization-account controls.
“train our models”https://www.anthropic.com/legal/consumer-terms
How long do you store my data?
Consumer deletion, model-improvement, policy-violation, classifier and feedback retention windows.
“within 30 days”https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data
Is my data used for model training? — consumer
Defines three consumer training paths: user allows it, safety review, or explicit opt-in.
“choose to allow”https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training
Use Incognito chats
Incognito avoids chat history/memory and training, but still carries default deletion and safety-policy handling.
“not used to improve Claude”https://support.claude.com/en/articles/12260368-use-incognito-chats
What is the Max plan?
Max 5x/20x multipliers, five-hour session reset and weekly/further discretionary caps.
“reset every five hours”https://support.claude.com/en/articles/11049741-what-is-the-max-plan
10.2
WBS/PBS 10.2
Team and Enterprise
Commercial privacy improves sharply for training, but organization-level visibility increases.
1Commercial Terms
No training on Customer Content by default.
2Persistent workspace
Chats/files exist so teams can return to them.
3Org governance
Primary Owner, SSO, audit, retention.
4Enterprise controls
Compliance API, custom retention, CMEK eligibility.
Privacy inversion: Team/Enterprise is more private from model training, but potentially less private from your employer. The Primary Owner can request exports that include conversations, files and usage patterns; Enterprise can expose chat/file data through Compliance API. SRC-ANT-015 SRC-ANT-016
6 rows
| Dimension | Team | Enterprise UI |
|---|---|---|
| Training | No by default | No by default |
| Retention | Saved until delete; ≤30d backend after delete | Default indefinite; custom ≥30d |
| Usage | Standard weekly / Premium 5× Standard | Usage-based plan: no plan/seat cap; admin spend caps |
| ZDR | No | No for normal Chat UI |
| Admin access | Primary Owner export | Primary Owner + Compliance API |
| Security | SSO/admin controls | Adds SCIM, audit, custom retention, IP/network controls, HIPAA-ready options |
Source packet for this tab 8 sources
Commercial Terms of Service
Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract.
“may not train models”https://www.anthropic.com/legal/commercial-terms
How long do you store my organization’s data?
Commercial API deletion window and persistent-work-product retention for Claude for Work/Enterprise.
“within 30 days”https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data
Is my data used for model training? — commercial
Commercial inputs/outputs are not training data by default; feedback or explicit choice can change that.
“will not use”https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training
What is the Enterprise plan?
Current usage-based Enterprise design: seat fee plus API-rate usage, no plan/seat cap, spend controls, compliance features.
“no per-seat usage limits”https://support.claude.com/en/articles/9797531-what-is-the-enterprise-plan
Who owns and manages the data of my team?
Explains employer/organization control: Primary Owner can manage and export member data including chats/files/usage.
“Primary Owner manages”https://support.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team
Access the Compliance API
Compliance API can export activity, chat and file content across supported deployments; some partner/cloud sessions are excluded.
“chat data”https://support.claude.com/en/articles/13015708-access-the-compliance-api
Configure custom data retention controls for Enterprise
Enterprise chat/project retention can be customized, but floor is 30 days and projects default to indefinite unless configured.
“minimum retention period is 30 days”https://support.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans
What is the Team plan?
Team Standard/Premium seat pricing, seat range and weekly usage limits.
“up to 150 seats”https://support.claude.com/en/articles/9266767-what-is-the-team-plan
10.3
WBS/PBS 10.3
API and Zero Data Retention
This is the strongest direct Anthropic data-retention regime, but it is feature- and model-scoped.
AClient
send request
BAnthropic API
eligible endpoint
CInference + safety
process request
DReturn response
stream/result
ENo content at rest
ZDR scope only
ZDR covers
Eligible Messages and Token Counting endpoints, API-key Claude Code and eligible Claude Code Enterprise, plus Claude Platform on AWS when enabled for the organization.
ZDR does not cover
Consumer products, normal Team/Enterprise product interfaces, Console/playground, many stateful features, third-party integrations, and Covered Models unless expressly authorized.
Residuals under ZDR: User Safety classifier results can remain; flagged content/legal holds can override normal retention; Access Transparency logs human views but automated safety processing is not a human-access event. SRC-ANT-010 SRC-ANT-011 SRC-ANT-019
Source packet for this tab 7 sources
Commercial Terms of Service
Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract.
“may not train models”https://www.anthropic.com/legal/commercial-terms
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Zero data retention agreement scope
Defines eligible products and the important safety-classifier residue that remains even in ZDR.
“does not store”https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to
Data retention practices for Covered Models
Covered frontier models impose 30-day safety retention across platforms, affecting ZDR arrangements.
“retained for 30 days”https://privacy.claude.com/en/articles/15425996-data-retention-practices-for-covered-models
Covered Models
Lists currently designated models and transition rules for ZDR/EFS.
“30-day data retention”https://support.claude.com/en/articles/15425695-covered-models
Access Transparency
Creates audit records of Anthropic human views of covered API/Claude Code data; automated processing is explicitly outside those events.
“Each human view”https://platform.claude.com/docs/en/manage-claude/access-transparency
Customer-managed encryption keys (CMEK)
Customer cloud KMS can control encryption keys for covered retained data, with explicit exclusions for telemetry/account data and short-lived caches.
“full control of the key”https://platform.claude.com/docs/en/manage-claude/cmek
20.1
WBS/PBS 20.1
Who can still see or process what?
Separate machine processing, retained content, human review, employer access and connector access.
Actor / system
Consumer
Team / Enterprise UI
API ZDR
AWS Bedrock none
GCP / Azure partner
Model-serving compute
YES — inference
YES — inference
YES — inference
YES — AWS-served inference
YES — partner/hosted inference
Anthropic automated safety
YES
YES
YES — ZDR does not disable safety
Model/provider rules vary; AWS does abuse detection
May apply; advanced models can add sharing
Anthropic human reviewer
Conditional safety/support
Conditional
Conditional safety/incident; Access Transparency can log
Normally NO content sharing; AWS review instead
Conditional on provider/model; Azure flags may reach Anthropic
Employer / Primary Owner
N/A unless managed account
YES — exports / Compliance API
Customer controls own API logs
Customer AWS admins/logs
Customer cloud admins/logs
Third-party connector
Only if enabled
Only if enabled / admin allowed
Outside ZDR if used
Only if app routes to it
Only if enabled
Cloud provider human review
N/A
Underlying infra not customer-facing
Depends Anthropic hosting
Conditional AWS review for models requiring it
Conditional Google/Azure safety operations
direct/normal processingconditional or configuration-dependentdocumented no-sharing path
Data that can exist even when prompt storage is minimized
account identitybillingusage metricsIP/device/log datasafety classifier resultsaudit eventsCompliance activityconnector-side copieslegal holds
These are not all stored in every product, and they do not all contain prompt text. The point is that “zero prompt/response retention” is narrower than “zero data about the interaction.”
Source packet for this tab 8 sources
Privacy Policy
Describes account, technical, input/output, integration, safety, legal-disclosure and third-party data processing categories.
“third-party services”https://www.anthropic.com/legal/privacy
Zero data retention agreement scope
Defines eligible products and the important safety-classifier residue that remains even in ZDR.
“does not store”https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to
Who owns and manages the data of my team?
Explains employer/organization control: Primary Owner can manage and export member data including chats/files/usage.
“Primary Owner manages”https://support.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team
Access the Compliance API
Compliance API can export activity, chat and file content across supported deployments; some partner/cloud sessions are excluded.
“chat data”https://support.claude.com/en/articles/13015708-access-the-compliance-api
Access Transparency
Creates audit records of Anthropic human views of covered API/Claude Code data; automated processing is explicitly outside those events.
“Each human view”https://platform.claude.com/docs/en/manage-claude/access-transparency
Customer-managed encryption keys (CMEK)
Customer cloud KMS can control encryption keys for covered retained data, with explicit exclusions for telemetry/account data and short-lived caches.
“full control of the key”https://platform.claude.com/docs/en/manage-claude/cmek
Government requests for user information
Anthropic describes legal-process standards, narrowing and notice practices; legal demands remain an exception to ordinary privacy promises.
“valid legal process”https://privacy.anthropic.com/en/articles/9519291-what-is-anthropic-s-policy-for-handling-governmental-requests-for-user-information
Amazon Bedrock data retention
Bedrock exposes account/project retention modes including guaranteed none where supported; models requiring review can retain inside AWS up to 30 days.
“guaranteed zero retention”https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html
20.2
WBS/PBS 20.2
Direct Claude data path
The same prompt can enter storage, training, safety, admin or connector paths depending on surface and settings.
DIA-20-002 · Direct Claude request path and side channelsdeterministic inline SVG
PATH 1Inference
Necessary to answer.
PATH 2Persistence
Chat history, files, stateful features.
PATH 3Safety
Automated classifiers; conditional review.
PATH 4External tools
Connector/provider’s own terms.
Source packet for this tab 9 sources
Consumer Terms of Service
Governs Claude Free/Pro/Max; permits model-improvement use unless user opts out, with feedback/safety exceptions and organization-account controls.
“train our models”https://www.anthropic.com/legal/consumer-terms
Commercial Terms of Service
Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract.
“may not train models”https://www.anthropic.com/legal/commercial-terms
Privacy Policy
Describes account, technical, input/output, integration, safety, legal-disclosure and third-party data processing categories.
“third-party services”https://www.anthropic.com/legal/privacy
How long do you store my data?
Consumer deletion, model-improvement, policy-violation, classifier and feedback retention windows.
“within 30 days”https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data
How long do you store my organization’s data?
Commercial API deletion window and persistent-work-product retention for Claude for Work/Enterprise.
“within 30 days”https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data
Is my data used for model training? — consumer
Defines three consumer training paths: user allows it, safety review, or explicit opt-in.
“choose to allow”https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training
Is my data used for model training? — commercial
Commercial inputs/outputs are not training data by default; feedback or explicit choice can change that.
“will not use”https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training
Use Incognito chats
Incognito avoids chat history/memory and training, but still carries default deletion and safety-policy handling.
“not used to improve Claude”https://support.claude.com/en/articles/12260368-use-incognito-chats
Use connectors to extend Claude’s capabilities
Third-party connectors run under separate infrastructure/terms and can create new data paths outside Anthropic retention boundaries.
“third-party services”https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities
20.3
WBS/PBS 20.3
AWS, Google Cloud and Microsoft Foundry
Partner clouds are not equivalent privacy wrappers around the same model.
DIA-20-003 · Same Claude family, materially different processor boundariesdeterministic inline SVG
4 rows
| Route | Who processes prompt/output? | Is Anthropic given content? | Zero-retention option | Covered/advanced-model caveat | Strongest privacy use |
|---|---|---|---|---|---|
| Anthropic API | Anthropic | Yes — first-party service execution | ZDR eligible endpoints | 30d Covered Model override unless authorized/EFS | Direct contract + ZDR/CMEK/Access Transparency |
| AWS Bedrock | AWS | AWS says NO sharing with model providers | data_retention_mode=none where eligible | Covered model may be retained/reviewed by AWS up to 30d | Strongest documented separation from Anthropic content |
| Google Agent Platform | General MaaS says no third-party sharing; specified Advanced AI can require Anthropic sharing | Conditional ZDR policy | Advanced AI logging up to 30d; some models require sharing | Google-governed workloads; verify model card/addendum | |
| Microsoft Foundry — Azure hosted | Azure infra + Anthropic independent processor role | Safety flags can be sent to Anthropic T&S | Arrangement-dependent | Safety/retention follows Anthropic terms/model rules | Azure residency + marketplace procurement |
Most important cloud distinction: Anthropic’s own AWS enterprise guidance describes Bedrock as a path where AWS stays the sole data processor. AWS separately states prompts/outputs are not shared with model providers. SRC-ANT-025 SRC-AWS-001
Source packet for this tab 10 sources
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Enterprise Transformation with Claude on AWS: Paths to Production
Anthropic describes Bedrock as AWS-served, AWS sole data processor, and highlights VPC-isolated workloads and Desktop-on-Bedrock paths.
“sole data processor”https://www.anthropic.com/webinars/enterprise-transformation-with-claude-on-aws-three-paths-to-production
Amazon Bedrock FAQs — Security
AWS states prompts/outputs are not shared with model providers, not used for base-model training, and can use PrivateLink.
“not shared with any model providers”https://aws.amazon.com/bedrock/faqs/
Amazon Bedrock data retention
Bedrock exposes account/project retention modes including guaranteed none where supported; models requiring review can retain inside AWS up to 30 days.
“guaranteed zero retention”https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html
Amazon Bedrock abuse detection
Covered Anthropic model traffic can be retained by AWS up to 30 days; flagged traffic can be reviewed by AWS, not shared with provider.
“human review performed by AWS”https://docs.aws.amazon.com/bedrock/latest/userguide/abuse-detection.html
Agent Platform zero data retention
Google documents conditions for ZDR: disable/request exceptions for logging, use store=false, and avoid features that necessarily retain data.
“achieving zero data retention”https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention
Agent Platform abuse monitoring
Advanced AI can impose 30-day prompt/response logging; some Anthropic models require data sharing with Anthropic for abuse monitoring.
“up to 30 days”https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/abuse-monitoring?hl=en
Partner models for Model-as-a-Service
Partner Claude uses managed Google endpoints, supports regional/multi-region controls, and Google states prompts/responses are not shared with third parties in the general MaaS path.
“managed API”https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/use-partner-models
Compare hosting options for Claude in Microsoft Foundry
Compares Azure-hosted and Anthropic-hosted Claude; Anthropic is seller/operator and an independent data processor in both paths.
“Hosted on Azure”https://learn.microsoft.com/azure/foundry/foundry-models/concepts/claude-models-hosting-comparison
Data, privacy, and security for Claude models in Microsoft Foundry
Azure-hosted data stays in selected Azure geography, but automatic safety flags can be sent to Anthropic Trust & Safety; Anthropic-hosted may process outside Azure.
“selected Azure geography”https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/claude-models/data-privacy
30.1
WBS/PBS 30.1
Retention windows
There is no single “Anthropic keeps data for X days” answer.
0ZDR response returned
eligible content not stored at rest
eligible content not stored at rest
≤30 ddeleted consumer/commercial backend
or Covered Model safety window
or Covered Model safety window
≤2 yflagged I/O
Usage Policy cases
Usage Policy cases
≤5 yconsumer training / feedback
if allowed / submitted
if allowed / submitted
≤6 ysome Compliance activity/session data
separate retention model
separate retention model
≤7 yT&S classifier scores
flagged sessions
flagged sessions
∞*Enterprise chats/projects
default until custom retention/delete
default until custom retention/delete
* “Indefinite” is the documented Enterprise default for chats/projects unless custom retention is configured; it is not a safety maximum.
Two apparent contradictions are both true: the general commercial policy says API I/O is automatically deleted within 30 days, while the newer feature-level API documentation says stateless conversation content is not retained by default. Treat 30 days as the commercial-policy outer bound where content is retained; use the specific feature documentation for the actual TTL. SRC-ANT-007 SRC-ANT-010
Source packet for this tab 9 sources
How long do you store my data?
Consumer deletion, model-improvement, policy-violation, classifier and feedback retention windows.
“within 30 days”https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data
How long do you store my organization’s data?
Commercial API deletion window and persistent-work-product retention for Claude for Work/Enterprise.
“within 30 days”https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Data retention practices for Covered Models
Covered frontier models impose 30-day safety retention across platforms, affecting ZDR arrangements.
“retained for 30 days”https://privacy.claude.com/en/articles/15425996-data-retention-practices-for-covered-models
Configure custom data retention controls for Enterprise
Enterprise chat/project retention can be customized, but floor is 30 days and projects default to indefinite unless configured.
“minimum retention period is 30 days”https://support.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans
Use Incognito chats
Incognito avoids chat history/memory and training, but still carries default deletion and safety-policy handling.
“not used to improve Claude”https://support.claude.com/en/articles/12260368-use-incognito-chats
Amazon Bedrock data retention
Bedrock exposes account/project retention modes including guaranteed none where supported; models requiring review can retain inside AWS up to 30 days.
“guaranteed zero retention”https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html
Agent Platform zero data retention
Google documents conditions for ZDR: disable/request exceptions for logging, use store=false, and avoid features that necessarily retain data.
“achieving zero data retention”https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention
Agent Platform abuse monitoring
Advanced AI can impose 30-day prompt/response logging; some Anthropic models require data sharing with Anthropic for abuse monitoring.
“up to 30 days”https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/abuse-monitoring?hl=en
30.2
WBS/PBS 30.2
Training is a separate switch from storage
A chat can be stored but never trained on; ZDR can eliminate storage but still leave safety metadata.
CONSUMERTraining allowed?
Current consumer pricing says opt-out.
COMMERCIALNo by default
Commercial Terms prohibit training on Customer Content.
FEEDBACKExplicit exception
Submitting feedback can place related conversation into research/training handling.
SAFETYPolicy exception
Flagged consumer data can be analyzed/trained for safeguards.
5 rows
| Case | Training? | Storage can still occur? | Source |
|---|---|---|---|
| Consumer, training ON | YES | YES — up to 5y de-identified training pipeline | SRC-ANT-006 SRC-ANT-008 |
| Consumer, training OFF | NO normal model improvement | YES — normal history/deletion/safety | SRC-ANT-006 SRC-ANT-008 |
| Commercial default | NO | YES — workspace/API feature retention | SRC-ANT-004 SRC-ANT-009 |
| Commercial feedback | Possible / explicit | YES — feedback context up to 5y | SRC-ANT-009 |
| API ZDR | NO by default | No content at rest, but safety/legal residue | SRC-ANT-010 SRC-ANT-011 |
Source packet for this tab 7 sources
Consumer Terms of Service
Governs Claude Free/Pro/Max; permits model-improvement use unless user opts out, with feedback/safety exceptions and organization-account controls.
“train our models”https://www.anthropic.com/legal/consumer-terms
Commercial Terms of Service
Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract.
“may not train models”https://www.anthropic.com/legal/commercial-terms
How long do you store my data?
Consumer deletion, model-improvement, policy-violation, classifier and feedback retention windows.
“within 30 days”https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data
Is my data used for model training? — consumer
Defines three consumer training paths: user allows it, safety review, or explicit opt-in.
“choose to allow”https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training
Is my data used for model training? — commercial
Commercial inputs/outputs are not training data by default; feedback or explicit choice can change that.
“will not use”https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Zero data retention agreement scope
Defines eligible products and the important safety-classifier residue that remains even in ZDR.
“does not store”https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to
30.3
WBS/PBS 30.3
Human, admin and audit access
Enterprise privacy depends on which human you are trying to be private from.
DIA-30-003 · The enterprise privacy inversion: less training, more employer visibilitydeterministic inline SVG
Anthropic humans
Access is conditional rather than normal product behavior. Eligible API customers can request Access Transparency, which emits events for human views under reason codes such as safety review or incident response. It does not record automated safety processing.
Your organization
Primary Owners can request exports containing member conversations/files/usage. Enterprise Compliance API can programmatically pull chats, files and activity across supported deployments.
Source packet for this tab 5 sources
Who owns and manages the data of my team?
Explains employer/organization control: Primary Owner can manage and export member data including chats/files/usage.
“Primary Owner manages”https://support.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team
Access the Compliance API
Compliance API can export activity, chat and file content across supported deployments; some partner/cloud sessions are excluded.
“chat data”https://support.claude.com/en/articles/13015708-access-the-compliance-api
Access Transparency
Creates audit records of Anthropic human views of covered API/Claude Code data; automated processing is explicitly outside those events.
“Each human view”https://platform.claude.com/docs/en/manage-claude/access-transparency
Customer-managed encryption keys (CMEK)
Customer cloud KMS can control encryption keys for covered retained data, with explicit exclusions for telemetry/account data and short-lived caches.
“full control of the key”https://platform.claude.com/docs/en/manage-claude/cmek
Anthropic Trust Center FAQ
States Claude for Work is multi-tenant and Anthropic does not offer single-tenancy; customer data is logically separated.
“multi-tenant architecture”https://trust.anthropic.com/faq
40.1
WBS/PBS 40.1
What is actually “capped” at each tier?
There are three different caps: usage capacity, normal content retention, and exceptional safety/feedback retention.
10 rows
| Surface | Price / seat | Usage cap | Training | Normal content retention | Training retention | Safety exception | ZDR? | Org/admin visibility | Regime |
|---|---|---|---|---|---|---|---|---|---|
| Free | $0 | Limited | Opt-out | Until delete; backend ≤30d after delete | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Pro | $20/mo or $200/y | Standard | Opt-out | Same consumer policy | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Max 5x | $100/mo | 5× Pro per session; 5h reset + weekly | Opt-out | Same consumer policy | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Max 20x | $200/mo | 20× Pro per session; 5h reset + weekly | Opt-out | Same consumer policy | ≤5y if training allowed | ≤2y flagged; scores ≤7y | No | No employer by default | Consumer |
| Team Standard | $20 annual / $25 monthly per seat | More than Pro; weekly | No training by default | Saved until delete; backend ≤30d after delete | No by default | ≤2y flagged; scores ≤7y | No | Primary Owner can export | Commercial |
| Team Premium | $100 annual / $125 monthly per seat | 5× Team Standard | No training by default | Same Team policy | No by default | ≤2y flagged; scores ≤7y | No | Primary Owner can export | Commercial |
| Enterprise UI | $20/seat/mo annual + API-rate usage | Usage-based: no plan/seat cap; spend caps | No training by default | Default indefinite; custom minimum 30d | No by default | Safety/legal exceptions | No (UI not ZDR) | Compliance API / exports | Commercial |
| Claude API — standard | API rates | Rate-limit tier / spend based | No training by default | Stateless conversation content not retained by default; outer policy ≤30d when retained | No by default | Safety/legal + feature state | No | Org API/admin logs | Commercial |
| Claude API — ZDR | API rates | Rate-limit tier / spend based | No training by default | No prompt/response at rest after response | No | Classifier/legal/safety residue | Yes, eligible endpoints | Org logs; Compliance data separate | Commercial |
| AWS Bedrock — retention none | AWS + model pricing | Quota based | No training | Zero retention where model/feature permits | No | Covered-model review rules can override | AWS retention mode | AWS org/IAM logs | Partner cloud |
Exactness warning
Anthropic does not publish a universal fixed message count for consumer plans because usage depends on model, context and workload. The stable published quantities are relative multipliers (Max 5×/20×; Team Premium 5× Standard), reset cadence, price, and Enterprise’s consumption-based design.
Source packet for this tab 5 sources
Plans & Pricing
Current consumer, Team and Enterprise prices, relative usage multipliers, model-training defaults and enterprise controls.
“Opt-out”https://claude.com/pricing
Choose a Claude plan
Consumer plan price and usage-capacity comparison.
“Limited”https://support.claude.com/en/articles/11049762-choose-a-claude-plan
What is the Enterprise plan?
Current usage-based Enterprise design: seat fee plus API-rate usage, no plan/seat cap, spend controls, compliance features.
“no per-seat usage limits”https://support.claude.com/en/articles/9797531-what-is-the-enterprise-plan
What is the Team plan?
Team Standard/Premium seat pricing, seat range and weekly usage limits.
“up to 150 seats”https://support.claude.com/en/articles/9266767-what-is-the-team-plan
What is the Max plan?
Max 5x/20x multipliers, five-hour session reset and weekly/further discretionary caps.
“reset every five hours”https://support.claude.com/en/articles/11049741-what-is-the-max-plan
50.1
WBS/PBS 50.1
What a €1T company can buy that a Free user cannot
The right answer is usually a privacy architecture, not merely an “Enterprise” subscription.
DIA-50-001 · Reference architecture for a €1T organization: do not use one privacy tier for everythingdeterministic inline SVG
Class
Recommended Claude route
Guardrail
P0 Public
Enterprise UI or API
Normal SSO + retention policy
P1 Internal
Enterprise UI
No consumer accounts; org controls
P2 Confidential
Enterprise + 30d retention or API standard
CMEK where useful; connectors allowlist
P3 Restricted
Eligible API ZDR / Bedrock retention none
DLP/pseudonymization, no stateful tools
P4 Crown jewel
Bedrock isolated path / EFS when eligible
Private network, customer keys, no logs containing payload
P5 Never external
Do not send to Claude
Local/on-prem model or deterministic tooling only
Enterprise contract/control plane
Commercial no-training default, custom retention ≥30 days for normal Enterprise chats/projects, SSO/SCIM, audit, Compliance API, HIPAA-ready options, CMEK eligibility, US-only inference, Access Transparency for eligible API content.
Restricted inference plane
Route high-sensitivity workloads through an internal gateway to eligible Anthropic API ZDR or AWS Bedrock retention-none. Disable non-ZDR stateful features, control connectors, keep payloads out of observability logs, and pseudonymize before transmission.
Do not assume “Enterprise Chat” means ZDR. Anthropic’s current API docs explicitly exclude the normal Team/Enterprise interfaces from ZDR; only specified API/Claude Code paths qualify. SRC-ANT-010
Source packet for this tab 14 sources
Commercial Terms of Service
Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract.
“may not train models”https://www.anthropic.com/legal/commercial-terms
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Zero data retention agreement scope
Defines eligible products and the important safety-classifier residue that remains even in ZDR.
“does not store”https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to
What is the Enterprise plan?
Current usage-based Enterprise design: seat fee plus API-rate usage, no plan/seat cap, spend controls, compliance features.
“no per-seat usage limits”https://support.claude.com/en/articles/9797531-what-is-the-enterprise-plan
Access the Compliance API
Compliance API can export activity, chat and file content across supported deployments; some partner/cloud sessions are excluded.
“chat data”https://support.claude.com/en/articles/13015708-access-the-compliance-api
Configure custom data retention controls for Enterprise
Enterprise chat/project retention can be customized, but floor is 30 days and projects default to indefinite unless configured.
“minimum retention period is 30 days”https://support.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans
Access Transparency
Creates audit records of Anthropic human views of covered API/Claude Code data; automated processing is explicitly outside those events.
“Each human view”https://platform.claude.com/docs/en/manage-claude/access-transparency
Customer-managed encryption keys (CMEK)
Customer cloud KMS can control encryption keys for covered retained data, with explicit exclusions for telemetry/account data and short-lived caches.
“full control of the key”https://platform.claude.com/docs/en/manage-claude/cmek
Data residency
Separates inference geography from workspace storage geography; global is default, US-only inference is configurable.
“Inference geo”https://platform.claude.com/docs/en/manage-claude/data-residency
Enable US-only inference for your organization
Enterprise setting pins inference—including background processing—to US servers at increased cost.
“within the United States”https://support.claude.com/en/articles/15422948-enable-us-only-inference-for-your-organization
Developing Enterprise Frontier Safeguards
EFS combines ZDR-like privacy with safety monitoring while storing monitoring data in customer-controlled cloud infrastructure.
“customer controls”https://www.anthropic.com/news/enterprise-frontier-safeguards
Enterprise Transformation with Claude on AWS: Paths to Production
Anthropic describes Bedrock as AWS-served, AWS sole data processor, and highlights VPC-isolated workloads and Desktop-on-Bedrock paths.
“sole data processor”https://www.anthropic.com/webinars/enterprise-transformation-with-claude-on-aws-three-paths-to-production
Amazon Bedrock FAQs — Security
AWS states prompts/outputs are not shared with model providers, not used for base-model training, and can use PrivateLink.
“not shared with any model providers”https://aws.amazon.com/bedrock/faqs/
Amazon Bedrock data retention
Bedrock exposes account/project retention modes including guaranteed none where supported; models requiring review can retain inside AWS up to 30 days.
“guaranteed zero retention”https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html
50.2
WBS/PBS 50.2
Is perfect privacy possible?
Absolute confidentiality from every external processor is incompatible with sending the secret to a remote model service.
NO
for absolute privacy from every external system while using a remote Claude service.
But several narrower goals are achievable
✓ No training on commercial content by default.
✓ No prompt/response stored at rest after eligible API ZDR responses.
✓ Anthropic-content isolation via Bedrock’s documented no-sharing model.
✓ Customer-controlled encryption keys for eligible retained data.
✓ Auditable Anthropic human access for eligible API surfaces.
✗ No guarantee that safety/metadata/legal pathways vanish.
✗ No single-tenant Claude for Work; Anthropic documents multi-tenancy.
DIA-50-002 · Why ZDR is not the same thing as perfect privacydeterministic inline SVG
6 rows
| Privacy objective | Achievable? | Best documented route | Residual |
|---|---|---|---|
| No model training | YES | Commercial Team/Enterprise/API default | Feedback/explicit opt-in exceptions |
| No content stored after response | YES, scoped | Eligible API ZDR / Bedrock retention none | Safety/legal/model/feature exceptions |
| Anthropic never receives prompt content | YES for some partner path | AWS Bedrock documents no sharing with model providers | AWS still processes content; Covered Model AWS retention may apply |
| No human vendor can ever review content | NO absolute guarantee | Minimize with ZDR / Bedrock / EFS | Safety/legal/incident pathways |
| No external processor at all | NO with remote Claude | Do not send data to Claude | Requires local/offline alternative |
| Single-tenant private Claude SaaS | NO documented Claude for Work option | Logical isolation / cloud partner architecture | Anthropic Trust Center says no single-tenancy |
Closest practical ceiling for highly confidential non-Covered-Model workloads: private internal gateway → DLP/pseudonymization → AWS Bedrock with retention mode
none where supported → regional endpoint → PrivateLink → customer KMS → payload logging disabled. For direct Anthropic: API ZDR + CMEK + Access Transparency + strict feature allowlist. For Covered Models, evaluate EFS/eligible transitional ZDR rather than assuming normal ZDR.Source packet for this tab 13 sources
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Zero data retention agreement scope
Defines eligible products and the important safety-classifier residue that remains even in ZDR.
“does not store”https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to
Data retention practices for Covered Models
Covered frontier models impose 30-day safety retention across platforms, affecting ZDR arrangements.
“retained for 30 days”https://privacy.claude.com/en/articles/15425996-data-retention-practices-for-covered-models
Covered Models
Lists currently designated models and transition rules for ZDR/EFS.
“30-day data retention”https://support.claude.com/en/articles/15425695-covered-models
Access Transparency
Creates audit records of Anthropic human views of covered API/Claude Code data; automated processing is explicitly outside those events.
“Each human view”https://platform.claude.com/docs/en/manage-claude/access-transparency
Customer-managed encryption keys (CMEK)
Customer cloud KMS can control encryption keys for covered retained data, with explicit exclusions for telemetry/account data and short-lived caches.
“full control of the key”https://platform.claude.com/docs/en/manage-claude/cmek
Anthropic Trust Center FAQ
States Claude for Work is multi-tenant and Anthropic does not offer single-tenancy; customer data is logically separated.
“multi-tenant architecture”https://trust.anthropic.com/faq
Developing Enterprise Frontier Safeguards
EFS combines ZDR-like privacy with safety monitoring while storing monitoring data in customer-controlled cloud infrastructure.
“customer controls”https://www.anthropic.com/news/enterprise-frontier-safeguards
Amazon Bedrock FAQs — Security
AWS states prompts/outputs are not shared with model providers, not used for base-model training, and can use PrivateLink.
“not shared with any model providers”https://aws.amazon.com/bedrock/faqs/
Amazon Bedrock data retention
Bedrock exposes account/project retention modes including guaranteed none where supported; models requiring review can retain inside AWS up to 30 days.
“guaranteed zero retention”https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html
Amazon Bedrock abuse detection
Covered Anthropic model traffic can be retained by AWS up to 30 days; flagged traffic can be reviewed by AWS, not shared with provider.
“human review performed by AWS”https://docs.aws.amazon.com/bedrock/latest/userguide/abuse-detection.html
Agent Platform abuse monitoring
Advanced AI can impose 30-day prompt/response logging; some Anthropic models require data sharing with Anthropic for abuse monitoring.
“up to 30 days”https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/abuse-monitoring?hl=en
Data, privacy, and security for Claude models in Microsoft Foundry
Azure-hosted data stays in selected Azure geography, but automatic safety flags can be sent to Anthropic Trust & Safety; Anthropic-hosted may process outside Azure.
“selected Azure geography”https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/claude-models/data-privacy
80.1
WBS/PBS 80.1
Master privacy matrix
A compact cross-product comparison for architecture decisions.
10 rows
| Surface | Primary processor | Training | Content storage | Deletion | ZDR | Key/network controls | Human review | Admin visibility | Analytical privacy posture |
|---|---|---|---|---|---|---|---|---|---|
| Consumer Free/Pro/Max | Anthropic | YES, unless opt-out / safety exception | Persistent until delete | 30d backend after deletion | No | No | Conditional | No | Low–medium |
| Consumer Incognito | Anthropic | NO | ≤30d | Auto | No | No | Conditional | No | Medium |
| Team | Anthropic | NO by default | Persistent until delete | ≤30d after delete | No | No | Conditional | Owner export | Medium |
| Enterprise UI | Anthropic | NO by default | Default indefinite; custom ≥30d | Custom | No (UI) | Eligible | Conditional | Compliance API | Medium–high vs Anthropic; low vs employer |
| API standard | Anthropic | NO by default | Not retained by default for stateless content; feature-specific | Feature-specific / outer ≤30d | No | Eligible | Conditional | Customer logs | High |
| API ZDR | Anthropic | NO | No prompt/response at rest after response | Immediate content non-storage | YES eligible | YES | Conditional; logged if Access Transparency | Customer logs | Very high from storage |
| Covered Model, normal | Anthropic/cloud | NO commercial | 30d safety retention | 30d | Overrides default ZDR unless special | Varies | Conditional | Varies | Reduced by safety override |
| AWS Bedrock retention none | AWS | NO | Zero where eligible | None | Provider-style zero retention | AWS KMS/PrivateLink | AWS conditional only | AWS admins | Very high from Anthropic |
| Google Agent Platform ZDR path | NO customer-data training | Zero if all ZDR conditions satisfied | Feature-specific | Conditional | CMEK/VPC-SC varies | Google conditional | GCP admins | High; Advanced AI caveats | |
| Microsoft Foundry Azure-hosted | Azure + Anthropic role | Commercial no training | Anthropic terms + Azure geo | Feature/model specific | Anthropic ZDR may apply by arrangement | Azure controls | Anthropic T&S exceptions | Azure admins | High but Anthropic remains processor |
Interactive exposure simulator
This is an analytical aid, not an Anthropic control. It converts documented configuration choices into an approximate exposure score (lower = less external content persistence/access).
Exposure index
—
0 best / 100 broadestSource packet for this tab 9 sources
Plans & Pricing
Current consumer, Team and Enterprise prices, relative usage multipliers, model-training defaults and enterprise controls.
“Opt-out”https://claude.com/pricing
How long do you store my data?
Consumer deletion, model-improvement, policy-violation, classifier and feedback retention windows.
“within 30 days”https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data
How long do you store my organization’s data?
Commercial API deletion window and persistent-work-product retention for Claude for Work/Enterprise.
“within 30 days”https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Who owns and manages the data of my team?
Explains employer/organization control: Primary Owner can manage and export member data including chats/files/usage.
“Primary Owner manages”https://support.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team
Access Transparency
Creates audit records of Anthropic human views of covered API/Claude Code data; automated processing is explicitly outside those events.
“Each human view”https://platform.claude.com/docs/en/manage-claude/access-transparency
Amazon Bedrock FAQs — Security
AWS states prompts/outputs are not shared with model providers, not used for base-model training, and can use PrivateLink.
“not shared with any model providers”https://aws.amazon.com/bedrock/faqs/
Agent Platform zero data retention
Google documents conditions for ZDR: disable/request exceptions for logging, use store=false, and avoid features that necessarily retain data.
“achieving zero data retention”https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention
Compare hosting options for Claude in Microsoft Foundry
Compares Azure-hosted and Anthropic-hosted Claude; Anthropic is seller/operator and an independent data processor in both paths.
“Hosted on Azure”https://learn.microsoft.com/azure/foundry/foundry-models/concepts/claude-models-hosting-comparison
90.1
WBS/PBS 90.1
Master deduplicated bibliography
41 deduplicated first-party sources, 51 fields each. Persistent source IDs are reused by every tab packet.
Copyright-safe excerpt policy: each source has three deliberately short verbatim fragments; full context is available through the linked source. No page is reproduced.
41 sources × 51 columns
| Source ID | Title | Publisher | Source type | Full-text URL | Published/updated | Checked | Authority tier | Scope | Plan/tier | Jurisdiction | Audience | Legal instrument | Processor role | Controller role | Training default | Training opt-out | Feedback handling | Content retention | Deletion SLA | Safety exception | Classifier retention | ZDR | Covered-model impact | EFS | Human access | Automated processing | Org/admin access | Third-party access | Metadata/telemetry | Encryption in transit | Encryption at rest | CMEK | Private network | Tenancy | Residency | Inference geo | Compliance API | Access Transparency | HIPAA/BAA | SOC/ISO | Usage/cap | Price | Seat range | Evidence status | Claim IDs | Research summary | Verbatim excerpt 1 | Verbatim excerpt 2 | Verbatim excerpt 3 | Notes |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SRC-ANT-001 | Plans & Pricing | Anthropic | Official pricing | https://claude.com/pricing | 2026-10-03 checked | 2026-10-03 | T1 — first-party authoritative | All plans | Consumer/Work | Global / contract dependent | Customers / admins / developers | — | — | — | Consumer opt-out; Work no training by default | — | — | See plan-specific sources | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Free limited; Max 5x/20x; Team Premium 5x Standard; Enterprise usage-based | — | — | [V] Verified official source | — | Current consumer, Team and Enterprise prices, relative usage multipliers, model-training defaults and enterprise controls. | Opt-out | 5x more usage | Custom data retention | — |
| SRC-ANT-002 | Choose a Claude plan | Anthropic | Help Center | https://support.claude.com/en/articles/11049762-choose-a-claude-plan | 2026-05-19 | 2026-10-03 | T1 — first-party authoritative | Consumer plans | Free/Pro/Max | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Free limited; Pro standard; Max 5x/20x | — | — | [V] Verified official source | — | Consumer plan price and usage-capacity comparison. | Limited | 5x Pro capacity | 20x Pro capacity | — |
| SRC-ANT-003 | Consumer Terms of Service | Anthropic | Legal terms | https://www.anthropic.com/legal/consumer-terms | Effective 2025-10-08 | 2026-10-03 | T1 — first-party authoritative | Consumer | Consumer | Global / contract dependent | Customers / admins / developers | Consumer Terms | — | — | Allowed unless opt-out; feedback/safety exceptions | — | — | — | — | — | — | — | — | — | — | — | Business-domain migration/admin controls possible | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Governs Claude Free/Pro/Max; permits model-improvement use unless user opts out, with feedback/safety exceptions and organization-account controls. | train our models | opt out | monitor and control | — |
| SRC-ANT-004 | Commercial Terms of Service | Anthropic | Legal terms | https://www.anthropic.com/legal/commercial-terms | Effective 2025-06-17 | 2026-10-03 | T1 — first-party authoritative | Commercial | Team/Enterprise/API | Global / contract dependent | Customers / admins / developers | Commercial Terms + DPA | — | — | No training on Customer Content by default | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract. | may not train models | Customer Content | need to know | — |
| SRC-ANT-005 | Privacy Policy | Anthropic | Privacy policy | https://www.anthropic.com/legal/privacy | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Cross-product | — | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Third-party integrations may receive Inputs/Outputs | IP/device/logs/integration data | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Describes account, technical, input/output, integration, safety, legal-disclosure and third-party data processing categories. | third-party services | IP address | Inputs and Outputs | — |
| SRC-ANT-006 | How long do you store my data? | Anthropic | Privacy Center | https://privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data | 2026-07-01 | 2026-10-03 | T1 — first-party authoritative | Consumer | Free/Pro/Max | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | Deleted chats ≤30d backend; training pipeline ≤5y; flagged I/O ≤2y; scores ≤7y; feedback 5y | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Consumer deletion, model-improvement, policy-violation, classifier and feedback retention windows. | within 30 days | up to 5 years | up to 7 years | — |
| SRC-ANT-007 | How long do you store my organization’s data? | Anthropic | Privacy Center | https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data | 2026-07-01 | 2026-10-03 | T1 — first-party authoritative | Commercial | Team/Enterprise/API | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | API auto-delete ≤30d outer bound; saved-work products persist; flagged I/O ≤2y; scores ≤7y | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Commercial API deletion window and persistent-work-product retention for Claude for Work/Enterprise. | within 30 days | consistent product experience | up to 2 years | — |
| SRC-ANT-008 | Is my data used for model training? — consumer | Anthropic | Privacy Center | https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training | 2026-03-16 | 2026-10-03 | T1 — first-party authoritative | Consumer | Free/Pro/Max | Global / contract dependent | Customers / admins / developers | — | — | — | Opt-out regime; safety and explicit opt-in exceptions | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Defines three consumer training paths: user allows it, safety review, or explicit opt-in. | choose to allow | flagged for safety review | explicitly opting in | — |
| SRC-ANT-009 | Is my data used for model training? — commercial | Anthropic | Privacy Center | https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training | 2026-08-18 | 2026-10-03 | T1 — first-party authoritative | Commercial | Team/Enterprise/API | Global / contract dependent | Customers / admins / developers | — | — | — | No by default; feedback/opt-in can be used | — | Related conversation up to 5y | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Commercial inputs/outputs are not training data by default; feedback or explicit choice can change that. | will not use | explicitly report feedback | up to 5 years | — |
| SRC-ANT-010 | API and data retention | Anthropic | Platform docs | https://platform.claude.com/docs/en/manage-claude/api-and-data-retention | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | API / Claude Code | API/ZDR | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | ZDR no prompt/response at rest after response; feature-specific exceptions | — | — | — | Messages + Token Counting eligible; many stateful features excluded | — | — | — | — | Compliance data may have separate 6y retention | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split. | not retained by default | does not store | six years | — |
| SRC-ANT-011 | Zero data retention agreement scope | Anthropic | Privacy Center | https://privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to | 2026-06-09 | 2026-10-03 | T1 — first-party authoritative | ZDR | API/Claude Code Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | Approval required; content storage eliminated for eligible paths | — | — | — | — | — | — | User Safety classifier results retained | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Defines eligible products and the important safety-classifier residue that remains even in ZDR. | does not store | User Safety classifier results | eligible Anthropic APIs | — |
| SRC-ANT-012 | Data retention practices for Covered Models | Anthropic | Privacy Center | https://privacy.claude.com/en/articles/15425996-data-retention-practices-for-covered-models | Policy effective 2026-06-09 | 2026-10-03 | T1 — first-party authoritative | Covered Models | Covered Models | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | 30d safety retention unless eligible special arrangement | — | — | — | Default ZDR override | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Covered frontier models impose 30-day safety retention across platforms, affecting ZDR arrangements. | retained for 30 days | on every platform | support our safety work | — |
| SRC-ANT-013 | Covered Models | Anthropic | Help Center | https://support.claude.com/en/articles/15425695-covered-models | 2026-09-01 | 2026-10-03 | T1 — first-party authoritative | Covered Models | Covered Models | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | 30d baseline | — | — | — | Eligible transitional ZDR/EFS paths exist | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Lists currently designated models and transition rules for ZDR/EFS. | 30-day data retention | real-time safety classifiers | customer controls | — |
| SRC-ANT-014 | What is the Enterprise plan? | Anthropic | Help Center | https://support.claude.com/en/articles/9797531-what-is-the-enterprise-plan | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Enterprise | Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | SSO/SCIM/audit/custom retention/Compliance API | — | — | — | — | — | — | — | — | — | — | — | — | — | No plan/seat cap on usage-based plan; admin spend caps | — | — | [V] Verified official source | — | Current usage-based Enterprise design: seat fee plus API-rate usage, no plan/seat cap, spend controls, compliance features. | no per-seat usage limits | standard API rates | spend limits | — |
| SRC-ANT-015 | Who owns and manages the data of my team? | Anthropic | Help Center | https://support.claude.com/en/articles/9265372-who-owns-and-manages-the-data-of-my-team | 2026-03-16 | 2026-10-03 | T1 — first-party authoritative | Team/Enterprise | Team/Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Primary Owner export/control | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Explains employer/organization control: Primary Owner can manage and export member data including chats/files/usage. | Primary Owner manages | conversations with Claude | usage patterns | — |
| SRC-ANT-016 | Access the Compliance API | Anthropic | Help Center | https://support.claude.com/en/articles/13015708-access-the-compliance-api | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Enterprise | Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Primary Owner can enable; org can pull content | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Compliance API can export activity, chat and file content across supported deployments; some partner/cloud sessions are excluded. | chat data | file content | across all your Claude deployments | — |
| SRC-ANT-017 | Configure custom data retention controls for Enterprise | Anthropic | Help Center | https://support.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Enterprise | Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | 30d minimum; default indefinite; project precedence | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Enterprise chat/project retention can be customized, but floor is 30 days and projects default to indefinite unless configured. | minimum retention period is 30 days | retained indefinitely | Project retention always takes precedence | — |
| SRC-ANT-018 | Use Incognito chats | Anthropic | Help Center | https://support.claude.com/en/articles/12260368-use-incognito-chats | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Claude apps | All Claude app plans | Global / contract dependent | Customers / admins / developers | — | — | — | No training | — | — | Auto-delete ≤30d unless enterprise override/safety | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Incognito avoids chat history/memory and training, but still carries default deletion and safety-policy handling. | not used to improve Claude | not saved to chat history | within 30 days | — |
| SRC-ANT-019 | Access Transparency | Anthropic | Platform docs | https://platform.claude.com/docs/en/manage-claude/access-transparency | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Eligible API customers | Eligible API customers | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | Human views logged under reason codes | Automated safety processing not logged as human access | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Creates audit records of Anthropic human views of covered API/Claude Code data; automated processing is explicitly outside those events. | Each human view | safety review | incident response | — |
| SRC-ANT-020 | Customer-managed encryption keys (CMEK) | Anthropic | Platform docs | https://platform.claude.com/docs/en/manage-claude/cmek | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Platform/Enterprise | Eligible Platform/Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Account data/telemetry and some caches excluded | — | — | Yes, eligible retained data | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Customer cloud KMS can control encryption keys for covered retained data, with explicit exclusions for telemetry/account data and short-lived caches. | full control of the key | audit logs | not encrypted under your key | — |
| SRC-ANT-021 | Data residency | Anthropic | Platform docs | https://platform.claude.com/docs/en/manage-claude/data-residency | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Platform API | — | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Inference global or US; workspace controls separately | — | — | — | — | — | — | — | — | [V] Verified official source | — | Separates inference geography from workspace storage geography; global is default, US-only inference is configurable. | Inference geo | Workspace geo | global | — |
| SRC-ANT-022 | Enable US-only inference for your organization | Anthropic | Help Center | https://support.claude.com/en/articles/15422948-enable-us-only-inference-for-your-organization | 2026-06-15 | 2026-10-03 | T1 — first-party authoritative | Enterprise | Usage-based Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | US-only inference available | — | — | — | — | — | — | — | — | [V] Verified official source | — | Enterprise setting pins inference—including background processing—to US servers at increased cost. | within the United States | background processing | increased cost | — |
| SRC-ANT-023 | Anthropic Trust Center FAQ | Anthropic | Trust Center | https://trust.anthropic.com/faq | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Commercial | Commercial | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Multi-tenant, logically separated | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | States Claude for Work is multi-tenant and Anthropic does not offer single-tenancy; customer data is logically separated. | multi-tenant architecture | do not offer single-tenancy | logically separated | — |
| SRC-ANT-024 | Developing Enterprise Frontier Safeguards | Anthropic | Product announcement | https://www.anthropic.com/news/enterprise-frontier-safeguards | 2026-09-01 | 2026-10-03 | T1 — first-party authoritative | EFS | Eligible enterprise / Covered Models | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | Phased rollout; customer-controlled cloud storage | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | EFS combines ZDR-like privacy with safety monitoring while storing monitoring data in customer-controlled cloud infrastructure. | customer controls | not Anthropic | zero data retention | — |
| SRC-ANT-025 | Enterprise Transformation with Claude on AWS: Paths to Production | Anthropic | Webinar | https://www.anthropic.com/webinars/enterprise-transformation-with-claude-on-aws-three-paths-to-production | 2026-09-16 | 2026-10-03 | T1 — first-party authoritative | AWS enterprise architectures | AWS enterprise | Global / contract dependent | Customers / admins / developers | — | AWS for Bedrock | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Anthropic describes Bedrock as AWS-served, AWS sole data processor, and highlights VPC-isolated workloads and Desktop-on-Bedrock paths. | sole data processor | stay inside your AWS boundary | VPC-isolated workloads | — |
| SRC-ANT-026 | Data Processing Addendum (DPA) guidance | Anthropic | Privacy Center | https://privacy.anthropic.com/en/articles/7996862-how-do-i-view-and-sign-your-data-processing-addendum-dpa | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Commercial | — | Global / contract dependent | Customers / admins / developers | DPA/SCC | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Commercial DPA and SCC path for customers needing processor/data-transfer commitments. | Data Processing Addendum | Commercial Terms | Standard Contractual Clauses | — |
| SRC-ANT-027 | Government requests for user information | Anthropic | Privacy Center | https://privacy.anthropic.com/en/articles/9519291-what-is-anthropic-s-policy-for-handling-governmental-requests-for-user-information | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Cross-product | — | Global / contract dependent | Customers / admins / developers | Law-enforcement/legal disclosure exception | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Anthropic describes legal-process standards, narrowing and notice practices; legal demands remain an exception to ordinary privacy promises. | valid legal process | legally prohibited | customer first | — |
| SRC-ANT-028 | Does Anthropic act as a data processor or controller? | Anthropic | Help Center | https://support.claude.com/en/articles/9267385-does-anthropic-act-as-a-data-processor-or-controller | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Commercial | — | Global / contract dependent | Customers / admins / developers | — | Anthropic processor for commercial customer content | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Clarifies commercial role split: customer generally controller and Anthropic processor for customer content. | data processor | data controller | commercial products | — |
| SRC-ANT-029 | Use connectors to extend Claude’s capabilities | Anthropic | Help Center | https://support.claude.com/en/articles/11176164-use-connectors-to-extend-claude-s-capabilities | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Connectors | — | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Connector-specific terms/infrastructure | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Third-party connectors run under separate infrastructure/terms and can create new data paths outside Anthropic retention boundaries. | third-party services | own infrastructure | data handling practices | — |
| SRC-ANT-030 | HIPAA-ready Enterprise plans | Anthropic | Help Center | https://support.claude.com/en/articles/13296973-hipaa-ready-enterprise-plans | 2026-07-23 | 2026-10-03 | T1 — first-party authoritative | Enterprise regulated use | Enterprise | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Eligible configuration | — | — | — | — | [V] Verified official source | — | Eligible Enterprise customers can configure a HIPAA-ready environment; feature coverage is not universal. | HIPAA-ready | Business Associate Agreement | eligible organizations | — |
| SRC-ANT-031 | Business Associate Agreements for commercial customers | Anthropic | Help Center | https://support.claude.com/en/articles/8114513-business-associate-agreements-baa-for-commercial-customers | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Enterprise/API regulated use | — | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | BAA for eligible commercial services | — | — | — | — | [V] Verified official source | — | Defines BAA eligibility and feature boundaries; third-party services are not automatically covered by Anthropic’s BAA. | Business Associate Agreement | Eligible Services | third-party | — |
| SRC-ANT-032 | What is the Team plan? | Anthropic | Help Center | https://support.claude.com/en/articles/9266767-what-is-the-team-plan | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Team | Team | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Standard weekly; Premium higher, pricing page says 5x Standard | — | — | [V] Verified official source | — | Team Standard/Premium seat pricing, seat range and weekly usage limits. | up to 150 seats | weekly limits | significantly more usage | — |
| SRC-ANT-033 | What is the Max plan? | Anthropic | Help Center | https://support.claude.com/en/articles/11049741-what-is-the-max-plan | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Max | Max | Global / contract dependent | Customers / admins / developers | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 5x or 20x Pro/session; 5h reset + weekly limits | — | — | [V] Verified official source | — | Max 5x/20x multipliers, five-hour session reset and weekly/further discretionary caps. | reset every five hours | five times | 20 times | — |
| SRC-AWS-001 | Amazon Bedrock FAQs — Security | AWS | AWS official docs | https://aws.amazon.com/bedrock/faqs/ | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Bedrock | — | Global / contract dependent | Customers / admins / developers | — | AWS | — | No customer I/O training | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | PrivateLink | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | AWS states prompts/outputs are not shared with model providers, not used for base-model training, and can use PrivateLink. | not shared with any model providers | AWS PrivateLink | not used to improve | — |
| SRC-AWS-002 | Amazon Bedrock data retention | AWS | AWS official docs | https://docs.aws.amazon.com/bedrock/latest/userguide/data-retention.html | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Bedrock | — | Global / contract dependent | Customers / admins / developers | — | AWS | — | — | — | — | none where eligible; aws_review up to 30d | — | — | — | — | — | — | AWS review where model requires | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Bedrock exposes account/project retention modes including guaranteed none where supported; models requiring review can retain inside AWS up to 30 days. | guaranteed zero retention | within the AWS boundary | up to 30 days | — |
| SRC-AWS-003 | Amazon Bedrock abuse detection | AWS | AWS official docs | https://docs.aws.amazon.com/bedrock/latest/userguide/abuse-detection.html | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Bedrock | — | Global / contract dependent | Customers / admins / developers | — | AWS | — | — | — | — | Covered model 30d | — | — | — | — | — | — | AWS conditional review | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Covered Anthropic model traffic can be retained by AWS up to 30 days; flagged traffic can be reviewed by AWS, not shared with provider. | human review performed by AWS | not shared | up to 30 days | — |
| SRC-GCP-001 | Agent Platform zero data retention | Google Cloud | GCP official docs | https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Agent Platform | — | Global / contract dependent | Customers / admins / developers | — | Google Cloud | — | — | — | — | — | — | — | — | Conditional; feature-specific | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Google documents conditions for ZDR: disable/request exceptions for logging, use store=false, and avoid features that necessarily retain data. | achieving zero data retention | do not enable | store = false | — |
| SRC-GCP-002 | Agent Platform abuse monitoring | Google Cloud | GCP official docs | https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/abuse-monitoring?hl=en | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Agent Platform | — | Global / contract dependent | Customers / admins / developers | — | Google Cloud; Anthropic sharing for specified advanced models | — | — | — | — | Advanced AI up to 30d | — | — | — | — | — | — | Google conditional review | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Advanced AI can impose 30-day prompt/response logging; some Anthropic models require data sharing with Anthropic for abuse monitoring. | up to 30 days | Authorized Google employees | sharing this data with Anthropic | — |
| SRC-GCP-003 | Partner models for Model-as-a-Service | Google Cloud | GCP official docs | https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/partner-models/use-partner-models | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Partner MaaS | — | Global / contract dependent | Customers / admins / developers | — | Google Cloud | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Region/multi-region options | — | — | — | — | — | — | — | — | [V] Verified official source | — | Partner Claude uses managed Google endpoints, supports regional/multi-region controls, and Google states prompts/responses are not shared with third parties in the general MaaS path. | managed API | data residency | not shared with third-parties | — |
| SRC-MS-001 | Compare hosting options for Claude in Microsoft Foundry | Microsoft | Microsoft Learn | https://learn.microsoft.com/azure/foundry/foundry-models/concepts/claude-models-hosting-comparison | Current 2026-10-03 | 2026-10-03 | T1 — first-party authoritative | Foundry | — | Global / contract dependent | Customers / admins / developers | — | Anthropic independent processor; Azure infra for Azure-hosted | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | Depends on hosting option | — | — | — | — | — | — | — | — | [V] Verified official source | — | Compares Azure-hosted and Anthropic-hosted Claude; Anthropic is seller/operator and an independent data processor in both paths. | Hosted on Azure | independent data processor | processed on Azure infrastructure | — |
| SRC-MS-002 | Data, privacy, and security for Claude models in Microsoft Foundry | Microsoft | Microsoft Learn | https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/claude-models/data-privacy | 2026-06-29 | 2026-10-03 | T1 — first-party authoritative | Foundry | — | Global / contract dependent | Customers / admins / developers | — | Azure + Anthropic roles | — | — | — | — | — | — | — | — | — | — | — | Anthropic T&S on flagged exceptions | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | [V] Verified official source | — | Azure-hosted data stays in selected Azure geography, but automatic safety flags can be sent to Anthropic Trust & Safety; Anthropic-hosted may process outside Azure. | selected Azure geography | exceptions-only basis | outside of Azure | — |
90.2
WBS/PBS 90.2
Method, confidence and update rules
Privacy claims are policy/version-sensitive. This report separates official facts from architectural inference and makes date scope explicit.
M1Authority first
Anthropic legal/docs/help, then AWS/GCP/Microsoft official docs.
M2Threat-model split
Anthropic vs employer vs cloud host vs connectors.
M3Retention ≠ training
Each axis independently verified.
M4Specific beats generic
Feature docs override broad summaries where scopes differ.
Evidence marks
[V] verified in official source.
[I] architectural inference from verified facts.
[D] dynamic commercial value — recheck before procurement.
Material exceptions to recheck
Covered Model designations, EFS rollout, pricing, usage quotas, cloud-provider retention modes, partner data-sharing addenda, HIPAA/BAA eligible-feature lists and regional availability.
Analytical index formula used only by simulator
base exposure by surface: consumer 52 / Enterprise UI 38 / API ZDR 12 / Bedrock none 8 + Covered/Advanced model: +25 + connectors enabled: +12 + feedback/explicit sharing: +15 clamped to 0..100 Interpretation: this is a transparent heuristic for comparing exposure pathways. It is not a security certification, risk probability, or vendor score.
Source packet for this tab 8 sources
Consumer Terms of Service
Governs Claude Free/Pro/Max; permits model-improvement use unless user opts out, with feedback/safety exceptions and organization-account controls.
“train our models”https://www.anthropic.com/legal/consumer-terms
Commercial Terms of Service
Commercial customers retain input rights and Anthropic does not train models on Customer Content under the default commercial contract.
“may not train models”https://www.anthropic.com/legal/commercial-terms
Privacy Policy
Describes account, technical, input/output, integration, safety, legal-disclosure and third-party data processing categories.
“third-party services”https://www.anthropic.com/legal/privacy
API and data retention
Current feature-level retention model, ZDR scope/exclusions, Covered Model override, Compliance API retention, partner-cloud processor split.
“not retained by default”https://platform.claude.com/docs/en/manage-claude/api-and-data-retention
Anthropic Trust Center FAQ
States Claude for Work is multi-tenant and Anthropic does not offer single-tenancy; customer data is logically separated.
“multi-tenant architecture”https://trust.anthropic.com/faq
Amazon Bedrock FAQs — Security
AWS states prompts/outputs are not shared with model providers, not used for base-model training, and can use PrivateLink.
“not shared with any model providers”https://aws.amazon.com/bedrock/faqs/
Agent Platform zero data retention
Google documents conditions for ZDR: disable/request exceptions for logging, use store=false, and avoid features that necessarily retain data.
“achieving zero data retention”https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention
Data, privacy, and security for Claude models in Microsoft Foundry
Azure-hosted data stays in selected Azure geography, but automatic safety flags can be sent to Anthropic Trust & Safety; Anthropic-hosted may process outside Azure.
“selected Azure geography”https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/claude-models/data-privacy