Anthropic Data Report Card

10/12 Policy grade · Claude · Free / Pro / Max

Agents from Anthropic

Graded on their own pages. They act in your name, so each gets Agent Safeguards as well as a Policy grade.

Agent	Agent Safeguardsout of 10	Policy gradegoverning policy	Note
Claude Cowork GAA sandbox on Anthropic’s servers, destroyed when the session ends	5/10	B10/12	Runs in a per-session cloud sandbox

The company grade stays about the account. A weak agent doesn’t quietly pull it down, and a strong company grade doesn’t vouch for an agent. All agents →

Reviewed Jun 2026 · some cells re-verified 9 Aug 2026 · policies are checked weekly for changes · stated — every cell is read from Anthropic's published policy; none has been tested (run by us) yet.

2Accessstated

Self-serve export of conversations plus account data from Settings → Privacy on the Claude web or desktop app (not available from iOS/Android), no fee.

Anthropic — export your Claude data · accessed Jun 2026 ↗

2Portabilitystated

The export is delivered as machine-readable JSON containing your messages, Claude's responses, conversation titles and timestamps.

Anthropic — export your Claude data · accessed Jun 2026 ↗

Connects to apps: No. Exporting your Claude data is a manual Settings → Privacy request (web and desktop only); the Anthropic API serves model inference, not programmatic retrieval of your own conversations — so an agent can't connect to fetch it for you. Anthropic — export your Claude data · accessed Jun 2026 ↗

1Frictionstated

Self-serve export with a "small delay" and a download link that expires after 24 hours; granular personal-data removal is handled by emailing privacy@anthropic.com rather than a button.

Anthropic — export your Claude data · accessed Jun 2026 ↗

2Deletionstated

Deleting a conversation removes it from view immediately and purges it from back-end systems within 30 days; self-serve account deletion excludes your data from future model training.

Anthropic — how long do you store my data? · accessed Jun 2026 ↗

2Sells / sharesstated

Anthropic states it does not sell your data and Claude is ad-free. Its Jul 2026 update adds third-party data-sharing rules only for when Claude executes multi-step tasks through connected apps — functional sharing to do the work you asked, not advertising or sale.

Anthropic — updates to our privacy policy · accessed Jun 2026 ↗

1Trains AIstated

Announced 28 Aug 2025 and effective 8 Oct 2025: the policy states new models are trained on Free / Pro / Max data "when this setting is on," and a clear account-wide opt-out exists, so this scores Moderate. Noteworthy: the rollout presented existing users a prominent "Accept" button with the training toggle pre-set to On in much smaller print — a consent-design pattern critics flagged as a dark pattern. Chats not opted out are retained up to 5 years (vs 30 days if opted out).

Anthropic — updates to consumer terms and privacy policy · accessed 9 Aug 2026 ↗

Source-verified Jun 2026 from Anthropic's own help pages and consumer terms, for the consumer tier (Claude Free / Pro / Max — Team, Enterprise and the API differ). All cells are "stated"; we have not yet run a live export or deletion to mark any cell "tested."

Events

Breaches, policy changes, and export-format changes we’ve recorded at Anthropic — other parties’ actions, alongside the enforcement record above. An event never changes a grade by itself. All events →

No events on record for Anthropic.

Grade history

Every change to this company’s grade since it was first published, and why. All grade changes →

9 Aug 2026Trains AI on your datasource

Score unchanged. Re-sourced to Anthropic's own privacy-policy update, which states consumer chats may be used to improve Claude “if you choose to allow it,” with no separate EEA opt-in. The dark-pattern characterisation in the previous note was an advocacy claim rather than a company statement and has been removed from this cell.

Anthropic — Privacy Policy update (eff. 28 Sep 2025) ↗

Previously: ~~Sonomos — is Claude GDPR compliant? 2026 guide~~

9 Aug 2026Trains AI on your datanote

Score unchanged. The note dated the training change to “Sep 2025”; the announcement is 28 Aug 2025, effective 8 Oct 2025. It now quotes Anthropic's own wording — models are trained on Free / Pro / Max data “when this setting is on” — rather than asserting a default the page does not state.

Anthropic — Updates to our consumer terms ↗

How to request your data from Anthropic

Go to https://privacy.claude.com/ — Anthropic's data-request entry point.
Claude → Settings → Privacy → Export data.
Expected wait: small delay (24-hour download link) — Anthropic's stated turnaround (we haven't independently timed a request yet).
No self-serve option, or want a formal request? Contact Anthropic — Privacy: mailto:privacy@anthropic.com.

These steps reflect Anthropic's documented process as of Jun 2026. The wait time is the company's stated figure, not yet independently measured by us.

Help me make a request →