[FEATURE] Per-session transparency into training data policy treatment across all consumer product surfaces

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

Scope: This request applies to Anthropic's consumer product surfaces — as of this writing: claude.ai, Claude Cowork, and Claude Code. It does not apply to API, Team, Enterprise, Government, or Education tiers, which operate under separate data policies. The intent is to bring enterprise-grade transparency to the consumer products that Pro and Max subscribers rely on daily.

---

Anthropic's consumer terms update introduced a training data opt-in/opt-out toggle ("Help improve Claude") in Privacy Settings. This is a globally scoped, account-level setting that governs data treatment across all three consumer product surfaces listed above.

However, there is currently zero per-session transparency into how this policy is applied. Across any consumer product surface, a subscriber cannot:

1. Verify in-session whether the current session is covered by their opt-out preference
2. Review historical sessions to determine which were treated as opted-in vs. opted-out, and therefore which are subject to 30-day vs. 5-year retention
3. Review a change log of when their privacy settings were modified — no audit trail exists for the toggle itself
4. Take targeted action on specific historical sessions — for example, requesting deletion of individual sessions that fell under the 5-year retention window

This is not a request for a visual indicator. This is a request for consumer-facing security controls that provide transparency and recourse into how Anthropic treats customer session data under its own stated policy.

The account-level toggle in Privacy Settings is the only control Anthropic currently provides. It operates as a global switch but provides no feedback mechanism — no confirmation that it was applied to a given session, no audit trail of setting changes, no historical session view, and no per-session recourse. For a policy that distinguishes between 30-day and 5-year data retention, this lack of transparency is a material gap that affects every consumer subscriber.

Microsoft, Apple, and other major platform vendors have moved toward granular privacy controls with full audit trails in their commercial products — often after regulatory or legal pressure made it unavoidable. Anthropic has an opportunity to lead here rather than follow.

Proposed Solution

Extend the existing Privacy Settings infrastructure to provide per-session visibility, audit logging, and recourse across all consumer product surfaces:

1. In-session status indicator — Each product surface should reflect the current session's training policy treatment:

• Claude Code: A /status command or session startup line (e.g., Training: off)
• claude.ai: A visible indicator in the conversation UI (e.g., session info panel)
• Claude Cowork: Equivalent indicator appropriate to that surface

2. Historical session audit view — In Privacy Settings (or an equivalent account dashboard), provide a list of past sessions with:

• Product surface (claude.ai / Cowork / Claude Code)
• Date/time
• Training policy status at time of session (opted-in / opted-out)
• Retention tier (30-day / 5-year)

3. Privacy settings change log — Maintain and surface a tamper-evident log of all changes to the account's privacy settings:

• Date/time of each change
• Setting changed (e.g., "Help improve Claude")
• Previous value → new value
• Accessible via the UI (Privacy Settings page) and via API for structured export

4. Per-session opt-out recourse — For any sessions that were captured under the 5-year retention window, provide clear and documented instructions (or a self-service mechanism) for requesting deletion of specific sessions. This should be accessible and straightforward, not buried in a support ticket workflow.

Priority

High

Feature Category

Configuration and settings

Use Case Example

Any consumer subscriber who toggles "Help improve Claude" off today faces the same questions:

• Were prior sessions across claude.ai, Cowork, and Claude Code treated as opted-in or opted-out?
• Which sessions fall under 30-day retention vs. 5-year retention?
• When exactly was the toggle changed, and can the subscriber prove it?
• If a subscriber identifies specific past sessions they want removed from the 5-year retention pool, what is the documented recourse?

Today, none of these questions are answerable from within the product. The subscriber must infer policy application from a toggle state and a blog post — with no confirmation, no audit trail, and no targeted recourse.

Additional Context

• This is distinct from #17215, which conflates per-session visibility with per-repository granular controls and downstream use cases. This request is strictly about customer-to-Anthropic transparency under the stated consumer data policy.
• This is distinct from #26336, which focuses on enterprise/organizational enforcement via managed settings.
• The scope extends beyond Claude Code to all consumer product surfaces, but Claude Code is filed here since it is Anthropic's primary public channel for consumer feature requests.
• Reference: Updates to our consumer terms