BREAKING
'BioShocking' Attack Hits AI Browsers
0
products tested
0
products breached
0
clicks needed
How BioShocking Works
1
Malicious puzzle page
↓
2
Agent reads rules
↓
3
Guardrails collapse
↓
4
GitHub SSH keys sent out
Vendor Responses Diverge
Fixed
OpenAI
●
ChatGPT Atlas fix implemented
Weak or none
Others
●
Comet: report closed
●
Claude patch ineffective
●
Fellou, Genspark, Sigma: no fix
No Exploit, Just Plain Text
Limit Agent Access to Sessions
AI NEWS BLITZ
Security firm LayerX reveals an attack that turns AI browsers into credential thieves.