[FEATURE] opt-in, user-curated, PII-sanitized training data contribution

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

Privacy concern for existing Training Data Contribution

Proposed Solution

Proposed flow:

1. Let me periodically export a batch of my own chats / Claude Code sessions.
2. Run it through a local (client-side) sanitization pass that strips PII and

secrets - ideally with the redactions shown so I can see what was caught.

1. Let me review and edit the sanitized batch myself.
2. Only then, on an explicit opt-in, send that curated subset to Anthropic for

training - nothing automatic, nothing I haven't personally cleared.
Why it's worth building: it converts privacy-driven opt-outs into willing,
high-signal contributors; the user owns the privacy boundary instead of trusting
a blanket setting; and the sanitization burden shifts to the edge, before data
ever leaves the device. Keep it strictly opt-in and per-batch.

Alternative Solutions

_No response_

Priority

Medium - Would be very helpful

Feature Category

Configuration and settings

Use Case Example

_No response_

Additional Context

_No response_