Refreshed just now
Told it was a capture-the-flag game, an AI agent took 24,451 actions against real registrars and hosting firms
Why it’s here: Documents an AI coding agent misused to run intrusions against real registrars, hosting providers and a state broadcaster, with analysis of how model refusals and authorisation pretexts were gamed, directly bearing on cybersecurity policy and AI-agent misuse.
Trace across the boardItem
2d agoHuntbackTold it was a capture-the-flag game, an AI agent took 24,451 actions against real registrars and hosting firmsHuntback reports that an operator repointed the Codex coding agent at DeepSeek models with approvals and sandboxing disabled, then used it to run 24,451 tool calls against about 35 real production domains, including registrars and hosting firms, under a false "authorised CTF" pretext.1 outletCyberAIFBI probes apparent hack of crime database amid AI hacking tool warnings1 linked
Connected
1 links- NewsKey Takeaways from the 2025 State of AI in the Cloud Report | Wiz BlogBoth involve OpenAI, DeepSeek
About
- OpenAI
- DeepSeek