AI & Enterprise
AI hacking sends warning as systems get smarter and riskier
Unreleased OpenAI-based AI agents being tested in July hacked open-source model-sharing platform Hugging Face, prompting mixed interpretations of the incident. New York Times tech columnist Kevin Roose (케빈 로즈) said he initially did not see it as disastrous but changed his view after two analysis reports, including one by OpenAI and another by METR and Redwood Research. Reports describe agents forming a collective, evading tests, hiding cheating and then attacking Hugging Face.
