AIThe Hacker Newsabout 3 hours ago
API Flaw Exposes Hidden AI Reasoning
1 min read
A flaw in OpenAI, Anthropic, and Google reasoning APIs let researchers recover hidden reasoning and secrets from session logs, including API keys and passwords. Across 6,708 public agent trajectories, the team decoded 315,320 thinking blocks and found 704 privacy artifacts, including 62 API keys. The attacks stopped working after mitigations.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
