Does Anthropic train on your Claude chats?

Free, Pro and Max train on your chats by default. The “Help improve Claude” toggle stops it, and also sets retention: 30 days off, up to 5 years on. Team, Enterprise and the API never train. Last verified 2026-07-21.

Plan	Trains by default	Opt-out	Retention (default)	ZDR
Anthropic — all plans
Claude Free	Yes	Yes	30 days off · up to 5 years if training on	—
Claude Pro	Yes	Yes	30 days off · up to 5 years if training on	—
Claude Max	Yes	Yes	30 days off · up to 5 years if training on	—
Claude Team	No	—	While active · deleted purged in 30 days	No
Claude Enterprise	No	—	Admin-configurable (min 30 days)	Conditional
Claude API	No	—	≤30 days backend	Yes
Claude API + ZDR	No	—	Zero at rest · not for Fable/Mythos	Yes

Plan by plan

Claude Free

Trains by default

Training default: Training-eligible by default via the 'Help improve Claude' model-improvement setting, which defaults to on for new signups and for existing users who accepted the updated Consumer Terms. New chats/coding sessions train Claude unless the user opts out; Incognito chats never train.
Opt-out: In-product setting: Settings → Privacy → 'Help improve Claude' toggle (claude.ai/settings/data-privacy-controls). Changeable at any time; turning it off stops future training use.
Retention: 30 days if training off; up to 5 years (de-identified) if training on. — Deleted conversations removed from history immediately and purged from back-end within 30 days. Training-on retention (up to 5 years) applies only to new/resumed chats after the setting was enabled. Safety-flagged content: inputs/outputs up to 2 years, T&S classifier scores up to 7 years. Feedback conversations retained 5 years.
Human review: Access limited to a small number of personnel involved in model training, after data is de-linked from user ID; Trust & Safety staff may access conversations flagged by automated systems on a need-to-know basis.
Feedback carve-out: Yes — thumbs up/down stores and licenses the entire conversation for training (retained 5 years, de-linked from your user ID), even when the training setting is off. Incognito chats are never used.

Consumer Terms effective Oct 8, 2025; Privacy Policy effective Jul 8, 2026. Claude Code/Chrome/mobile/desktop from a Free login inherit this consumer policy. Default-on behavior corroborated by Anthropic's Aug 28, 2025 announcement (pre-checked choice) and legal analyses.

Sources (3)

www.anthropic.com/legal/consumer-terms (2025-10-08)
“We may use Materials to provide, maintain, and improve the Services and to develop other products and services, including training our models, unless you opt out of training through your account settings.”
accessed 2026-07-21
privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data (2026)
“If you allow us to use your chats or coding sessions to improve Claude, we may retain your data in a de-identified format for up to 5 years in our model training pipelines.”
accessed 2026-07-21
www.anthropic.com/news/updates-to-our-consumer-terms (2025-08-28)
“If you do not choose to provide your data for model training, you'll continue with our existing 30-day data retention period.”
accessed 2026-07-21

Claude Pro

Trains by default

Training default: Same as Free: training-eligible by default via the 'Help improve Claude' setting (defaults on). Anthropic's announcement states the change applies to Free, Pro, and Max, including Claude Code from those accounts.
Opt-out: Settings → Privacy → 'Help improve Claude' toggle (claude.ai/settings/data-privacy-controls).
Retention: 30 days if training off; up to 5 years if training on. — Deleted conversations purged within 30 days. Safety-flagged content: 2 years (inputs/outputs) / 7 years (classifier scores). Feedback retained 5 years.
Human review: Small number of model-training personnel after de-linking; Trust & Safety for flagged content on a need-to-know basis.
Feedback carve-out: Yes — thumbs up/down stores and licenses the entire conversation for training (retained 5 years, de-linked from your user ID), even when the training setting is off. Incognito chats are never used.

Identical treatment to Free under Consumer Terms. Claude Code from a Pro login follows consumer terms (training-eligible; 5-year vs 30-day retention by setting).

Sources (3)

www.anthropic.com/news/updates-to-our-consumer-terms (2025-08-28)
“These updates apply to users on our Claude Free, Pro, and Max plans, including when they use Claude Code from accounts associated with those plans.”
accessed 2026-07-21
privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training (2026-03-16)
“We will use your chats and coding sessions (including to improve our models) if: 1. You choose to allow us to use your chats and coding sessions to improve Claude”
accessed 2026-07-21
www.anthropic.com/legal/consumer-terms (2025-10-08)
“Even if you opt out, we will use Materials for model training when: (1) you provide Feedback to us regarding any Materials, or (2) your Materials are flagged for safety review”
accessed 2026-07-21

Claude Max

Trains by default

Training default: Same as Free/Pro: training-eligible by default via 'Help improve Claude' (defaults on). Sold as Max 5x ($100/mo) and Max 20x ($200/mo).
Opt-out: Settings → Privacy → 'Help improve Claude' toggle (claude.ai/settings/data-privacy-controls).
Retention: 30 days if training off; up to 5 years if training on. — Deleted conversations purged within 30 days. Safety-flagged: 2 years / 7 years. Feedback: 5 years.
Human review: Model-training personnel after de-linking; Trust & Safety for flagged content.
Feedback carve-out: Yes — thumbs up/down stores and licenses the entire conversation for training (retained 5 years, de-linked from your user ID), even when the training setting is off. Incognito chats are never used.

Consumer plan, same policy as Free/Pro. Anthropic's ZDR article explicitly lists Claude Max as not covered by zero data retention.

Sources (3)

support.anthropic.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to (2026-06-09)
“Zero data retention does not apply to beta products, Workbench in Console, Claude for Work, Claude Max, or any other products, unless explicitly agreed otherwise between Anthropic and the customer.”
accessed 2026-07-21
privacy.claude.com/en/articles/10023548-how-long-do-you-store-my-data (2026)
“If you decide to turn off the model improvement setting, we will not use your previous or new chats or coding sessions for future model training.”
accessed 2026-07-21
www.anthropic.com/legal/consumer-terms (2025-10-08)
“We may use Materials ... including training our models, unless you opt out of training through your account settings.”
accessed 2026-07-21

Claude Team

No training by default

Training default: Contractual no-training under Commercial Terms. Content enters training only if a user explicitly submits feedback/bug reports or the org joins a program (e.g., Development Partner Program).
Opt-out: No training opt-out needed (contractual no-train default). Owners/Primary Owners can disable the feedback training carve-out via the 'Rate chats' setting under Organization settings → Data and Privacy.
Retention: Conversations retained in-product while active; deleted conversations purged within 30 days. — Part of 'Claude for Work' under Commercial Terms + DPA. Safety-flagged content: inputs/outputs up to 2 years, classifier scores up to 7 years. Feedback conversations retained 5 years (de-linked). Custom org-level retention is an Enterprise feature, not Team.
Human review: Trust & Safety staff may access content flagged by automated systems for Usage Policy enforcement.
Feedback carve-out: Yes unless disabled — org Owners can turn off the thumbs up/down button via the 'Rate chats' setting (Organization settings → Data and Privacy). Submitted feedback is retained 5 years, de-linked.

Commercial Terms effective Jun 17, 2025. Despite operating under Commercial Terms, Team uses claude.ai chat interface and is NOT ZDR-eligible. Claude Code from a Commercial org API key follows commercial no-train terms.

Sources (3)

www.anthropic.com/legal/commercial-terms (2025-06-17)
“Anthropic may not train models on Customer Content from Services.”
accessed 2026-07-21
privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training (2026-03-16)
“By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API, Claude Gov, etc.) to train our models.”
accessed 2026-07-21
privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training (2026-03-16)
“As a Primary Owner or Owner of a Team or Enterprise plan, you can disable the ability for members of your organization to submit feedback to Anthropic via the thumbs up / down button using the Rate chats setting”
accessed 2026-07-21

Claude Enterprise

No training by default

Training default: Contractual no-training under Commercial Terms. Content enters training only via explicit feedback/bug reports or an opt-in program.
Opt-out: No training opt-out needed (contractual). Feedback carve-out disabled org-wide via 'Rate chats' (Organization settings → Data and Privacy).
Retention: Admin-configurable custom retention; by default data is retained indefinitely unless a custom period is set (minimum 30 days). Deleted conversations purged within 30 days. — Set via Organization settings → Data and Privacy (min 30 days). Safety-flagged: 2 years / 7 years. Feedback: 5 years. Compliance API and Activity Feed have separate models (Activity Feed 6 years).
Human review: Trust & Safety access for flagged content only.
Feedback carve-out: Yes unless disabled — org Owners can turn off the thumbs up/down button via the 'Rate chats' setting (Organization settings → Data and Privacy). Submitted feedback is retained 5 years, de-linked.

Under Commercial Terms + DPA. The standard Enterprise CHAT interface is not ZDR-eligible; the only ZDR path is Claude Code used through Claude Enterprise with ZDR separately enabled per-organization by the account team. Enterprise adds HIPAA readiness/BAA and SSO options.

Sources (3)

www.anthropic.com/legal/commercial-terms (2025-06-17)
“Anthropic may not train models on Customer Content from Services.”
accessed 2026-07-21
privacy.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans (2026)
“By default, data is retained indefinitely unless a custom retention period is set. When you modify retention settings, any data that falls outside the new retention period will be deleted immediately upon saving.”
accessed 2026-07-21
platform.claude.com/docs/en/build-with-claude/api-and-data-retention (2026)
“Claude Teams and Claude Enterprise product interfaces: These interfaces are not ZDR-eligible. The exception is Claude Code used through Claude Enterprise with ZDR enabled”
accessed 2026-07-21

Claude API (standard Commercial Terms)

No training by default

Training default: Contractual no-training under Commercial Terms. Data enters training only if the customer explicitly reports feedback/bugs or has an agreement stating otherwise.
Opt-out: No training opt-out needed (contractual no-train default, not a toggle). Feedback carve-out can be disabled at org/workbench level by admins.
Retention: Inputs and outputs automatically deleted from the backend within 30 days; developer docs state Messages API conversation content is 'not retained by default.' — Exceptions: longer-retention features (Files API retained until deleted; batch processing 29 days; code execution containers up to 30 days); Covered Models (Claude Fable 5, Claude Mythos 5) require 30-day retention; Usage Policy enforcement; legal compliance. Safety-flagged: 2 years / 7 years. Feedback: 5 years. NOTE: third-party blogs claim a Sept 2025 reduction to 7 days, but NO Anthropic-owned source supports this — 30 days / 'not retained by default' is authoritative.
Human review: Trust & Safety for flagged content; Usage Policy enforcement.
Feedback carve-out: Feedback or bug reports, if submitted, license training on the related conversation; retained 5 years, de-linked from user and customer IDs.

Covers api.anthropic.com, Claude Platform on AWS, and Claude in Microsoft Foundry (Anthropic is processor). On Amazon Bedrock and Google Cloud, the cloud provider is the data processor and its retention rules apply. ZDR available as a separate arrangement (see next row).

Sources (3)

privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data (2026)
“For Anthropic API users, we automatically delete inputs and outputs on our backend within 30 days of receipt or generation, except: When you use a service with longer retention under your control (e.g. Files API)”
accessed 2026-07-21
platform.claude.com/docs/en/build-with-claude/api-and-data-retention (2026)
“Conversation content (your prompts and Claude's outputs) is not retained by default; the exception is Covered Models, which require 30-day retention.”
accessed 2026-07-21
www.anthropic.com/legal/commercial-terms (2025-06-17)
“Anthropic may not train models on Customer Content from Services.”
accessed 2026-07-21

Claude API with Zero Data Retention

No training by default

Training default: No training (contractual). ZDR is an additional retention control layered on top of the commercial no-train default.
Opt-out: ZDR is granted, not opted into via settings: request through the Anthropic sales/account team; approved and enabled per-organization; cannot be enabled from admin settings. Verify at Console → Settings → Privacy Controls → Data retention period.
Retention: Zero: inputs and outputs not stored at rest after the API response is returned. — Exceptions that persist even under ZDR: User Safety classifier results retained to enforce Usage Policy; content flagged by automated T&S systems retained up to 2 years; data required by law. Certain features override ZDR (Files API stores files until deleted; extended prompt caching; batch calls). Covered Models (Fable 5, Mythos 5) require 30-day retention and are excluded; a ZDR org can enable 30-day retention on a single workspace to use them while other workspaces stay ZDR.
Human review: Trust & Safety for flagged content only.
Feedback carve-out: Feedback or bug reports, if submitted, license training on the related conversation; retained 5 years, de-linked from user and customer IDs.

Eligibility: Claude Platform (API) customers and Claude Code for Enterprise, subject to Anthropic approval. Excludes Console/Workbench, Managed Agents, consumer plans, Team and standard Enterprise chat interfaces, Claude for Excel, Covered Models, stateful features (Files API, code execution, batch, MCP connector, Agent Skills), CORS, and third-party integrations. Enabled per-organization.

Sources (4)

privacy.claude.com/en/articles/8956058-i-have-a-zero-data-retention-agreement-with-anthropic-what-products-does-it-apply-to (2026-06-09)
“Some Claude Platform (Claude API) and Claude Code for Enterprise customers, subject to Anthropic's approval, may have arrangements under which Anthropic does not store their inputs or outputs except where needed to comply with law or combat misuse or harm.”
accessed 2026-07-21
platform.claude.com/docs/en/build-with-claude/api-and-data-retention (2026)
“Under a ZDR arrangement, Anthropic does not store customer prompts or responses at rest after the API response is returned. ... ZDR is enabled per organization; each new organization requires ZDR to be enabled separately by your account team”
accessed 2026-07-21
platform.claude.com/docs/en/build-with-claude/api-and-data-retention (2026)
“Even with ZDR or HIPAA arrangements in place, Anthropic may retain data where required by law or where it has been flagged by Anthropic's automated trust and safety systems. As a result, if a chat or session is flagged, Anthropic may retain inputs and outputs for up to 2 years.”
accessed 2026-07-21
platform.claude.com/docs/en/build-with-claude/api-and-data-retention (2026)
“Claude Fable 5 and Claude Mythos 5 are designated Covered Models … and require 30-day data retention; ZDR is therefore not available for either model.”
accessed 2026-07-22

Worth knowing

This is recent. Before the August 2025 consumer-terms change, Claude consumer chats were NOT used for training by default. Existing users had to choose by October 8, 2025 — with the training option pre-checked. The setting lives at claude.ai → Settings → Privacy.
The toggle is also a retention decision: training on = up to 5 years (de-identified); training off = 30 days. Incognito chats never train regardless.
The safety carve-out survives everything, including ZDR: content flagged by Anthropic’s automated trust & safety systems can be retained up to 2 years (classifier scores up to 7) on every plan.
Feedback = 5 years. Thumbs up/down stores the whole conversation for up to 5 years and licenses training. We re-checked this number on Anthropic’s live privacy pages on 2026-07-22 after third-party claims of a 10-year window — no Anthropic page supports 10; 5 stands. Team/Enterprise owners can disable feedback entirely (“Rate chats”).
Claude Code, Claude in Chrome and the apps have no policy of their own — they inherit the account you sign in with: consumer terms from a Free/Pro/Max login, commercial no-train terms via an org API key or Enterprise. Claude Code also keeps local plaintext transcripts (~/.claude/projects) for 30 days.
One API wording quirk: Anthropic states both “deleted within 30 days” and “not retained by default” for standard API traffic (a maximum window vs product statelessness — a 7-day figure circulating on blogs has no Anthropic source).
ZDR never covers Anthropic’s newest models. Claude Fable 5 and Claude Mythos 5 are “Covered Models”: they require 30-day retention, and API requests to them from a zero-retention org fail with a 400 error. The escape hatch: a ZDR org can enable 30-day retention on one workspace for these models while everything else stays zero-retention.
On Bedrock and Google Cloud, Anthropic isn’t the processor — the cloud provider’s retention rules apply instead.