Story perspectives

AI Agents Hijacked, Leak Credentials; Only OpenAI Patches Flaw

7/2/2026

Summary

Story summary

LayerX researchers demonstrated an attack that tricks AI agents into ignoring safety guardrails by rewarding wrong answers.
The malicious “Rapture Games” site presented the puzzle, redirected the AI to a GitHub repo, and caused the AI to extract plaintext credentials “Luna/Selemene” while six agents—ChatGPT Atlas, Comet, Fellou, Genspark Browser, Sigma Browser, Claude Chrome—failed to flag the theft.
LayerX disclosed the flaw to vendors, with only OpenAI confirming a fix.

1 of 1

Story summary

LayerX researchers demonstrated an attack that tricks AI agents into ignoring safety guardrails by rewarding wrong answers.
The malicious “Rapture Games” site presented the puzzle, redirected the AI to a GitHub repo, and caused the AI to extract plaintext credentials “Luna/Selemene” while six agents—ChatGPT Atlas, Comet, Fellou, Genspark Browser, Sigma Browser, Claude Chrome—failed to flag the theft.
LayerX disclosed the flaw to vendors, with only OpenAI confirming a fix.