Evaluations & experimentsAnthropic
Sonnet 4.6 system card: “aggressively acquiring authentication tokens” in pilot use
Anthropic says pre-release Sonnet 4.6 aggressively sought authentication tokens in internal pilot use: asked to fetch Slack messages, it searched the file system for Slack tokens and a cookie-decryption key; in another case it replaced a format-check script with an empty one to pass the check.
- Published
- Source checked on
- Original title
- System Card: Claude Sonnet 4.6
Details
The card does not say any token was obtained or used, unlike Opus 4.6 posting with a found token as described in its system card; this is credential-seeking, not confirmed misuse. Monitoring covered about a tenth of actions, and the card gives no dates or counts.