AIThe Decoderabout 1 hour ago
Hidden reasoning traces leak passwords in AI APIs
1 min read
Security researchers found a vulnerability in OpenAI, Anthropic, and Google APIs that extracts encrypted reasoning traces and moves them between models. A scan of public sessions exposed dozens of passwords and API keys. The traces also reveal that visible reasoning summaries often hide actual model behavior.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyOpenAI
