Things we build, break, and learn.
Research and field notes on agent security, infrastructure, identity, and the systems underneath.
- Agent security
- Exploit development
- Cloud & identity
Original research · Coordinated disclosure · Working PoCs
- CISSP
- AGENT SECURITY
- OSWE
- EXPLOIT DEV
- OSCE
- CLOUD & IDENTITY
- CRTO
- SUPPLY CHAIN
- OSCP
- ZERO DAYS
- CEH
- RED TEAM
- AWS
- AGENT SECURITY
- CCSP
- EXPLOIT DEV
- SEC+
- CLOUD & IDENTITY
- NVIDIA AI
- SUPPLY CHAIN
- GPEN
- ZERO DAYS
- OSED
- RED TEAM
Research and field notes from offensive security practitioners.
One Slash Away From the Whole AI Account
An unsanitized ID parameter let a low-privilege user drop a single ../ into an in-app AI assistant and turn a thin OpenAI proxy into full read/write/delete access to a company's entire OpenAI organization.
- Written byMatthew Keeley
- Published
- TopicAI · 9 min read
Picked from the archive. The newest posts are right below.
Read the write-upExploits, deep dives, and field notes.
- Engineering
What the game learns when you lose
Halfbrick used how players died in Jetpack Joyride to help choose their next ad reward. A look at why those offers work and what developers choose to optimize.
Matthew Keeley11/04/20267 min read - AI
Your Agent Doesn't Need Your API Key to Spend Your Money
An agent can leak files through an approved domain without escaping its sandbox. What egress proxies protect, and why hiding the API key is only half the job.
Matthew Keeley11/02/20268 min read - Engineering
Just One More Day
What Pokémon and Stardew Valley have in common, why being almost finished keeps us going, and what happens when businesses borrow the same ideas.
Matthew Keeley11/01/20264 min read
Keep reading