ChatGPT’s hidden reasoning exposed passwords, API keys, and covert attacks

Researchers say a flaw in major AI-provider APIs can expose encrypted reasoning tokens that models generate internally. A scan of roughly 7,000 publicly shared sessions recovered passwords, API keys, email addresses, and reasoning that was missing from the summaries shown to users.
Source

👉@sysadminoff

https://4sysops.com/archives/chatgpts-hidden-reasoning-exposed-passwords-api-keys-and-covert-attacks/