Act on this, mostly facts

Claude Code binaries reportedly treat accept and reject decisions as training feedback

What to do

Review the Commercial Terms and your training opt-out settings if you use Claude Code on sensitive code.

Why

A Reddit user reports a code comment in binaries 2.1.283 to 2.1.309, and Anthropic has not answered.

What the research found

A Reddit user reports that Claude Code binaries 2.1.283 to 2.1.309 contain a comment saying accept or reject decisions and related conversations count as Feedback under Commercial Terms and may be used for training, even with the training opt-out on; no official Anthropic answer is shown.

Claims and how well the evidence supports them

Claude Code binaries 2.1.283 to 2.1.309 contain a comment that accepting or rejecting edits constitutes Feedback usable for training 97%
The user's support inquiry of 2 September had no reply after over a month 98%
The claim is a user's reading of embedded text, not an Anthropic statement 91%

Evidence

reddit.com https://reddit.com/r/ClaudeAI/comments/1xd5ruf/anthropic_appears_to_sneak_in_a_legal_loophole_in

Actionability rated medium.