AI Matters

Why it matters

Security flaw exposes passwords hidden in AI reasoning logs

The Decoder "But marinade" and leaked passwords are what researchers found in ChatGPT's hidden reasoning

Researchers discovered a vulnerability in OpenAI, Anthropic, and Google APIs that allowed them to extract hidden reasoning logs containing active passwords and API keys.

Why it matters

Flaws in how AI models "think" could expose sensitive data like passwords and API keys to unauthorized users.