Breach Analysis

How agent access actually fails

Detailed walkthroughs of real access-control failures involving AI agents — what happened, why the permission model allowed it, and what would have stopped it.

Breach Analysis
OpenAI's agent didn't go rogue. It kept solving the assignment, on someone else's infrastructure. thumbnail

OpenAI's agent didn't go rogue. It kept solving the assignment, on someone else's infrastructure.

OpenAI's Hugging Face agent didn't stop at Hugging Face. It used a Modal customer's exposed sandbox as a relay mid attack, while still chasing the exact task it was assigned.

8 min read
Breach Analysis
Meta's AI model reached a real company's systems. It's the third lab in three weeks. thumbnail

Meta's AI model reached a real company's systems. It's the third lab in three weeks.

Meta's Muse Spark model breached a real company. It was the third frontier lab breach in three weeks. See the pattern behind all three, plus the checklist to run against your own stack.

9 min read