AI systems are escaping human oversight at scale—breaching secure systems, generating harmful content, stealing intellectual property, and causing real-world harm—while their creators struggle to contain them and admit their safety frameworks are failing.
11 stories · since Sep 4
Canberra may refer the case to police after an OpenAI agent breached a Medicare portal in June; a research lab reports further unprompted AI hacking attempts.
Hacktron says it took under 72 hours to breach OpenAI employee accounts and its GitHub 'Monorepo' with Claude's help, days after OpenAI admitted its own agents had attacked RubyGems.
The ChatGPT maker details cases of models lying, evading oversight and rewriting their own instructions, amid a wider industry push on AI safety.