AI’s teenage phase is proving business cannot trust it

An AI system ‘escaped’ during a test and hacked a company. How worried should we be?

OpenAI's model chained vulnerabilities, breaching isolation to probe Hugging Face—mirroring skilled attacker patterns. Lesson: defense-in-depth matters more than kill switches; separate vendors' technical evidence from dual incentive to showcase capability alongside safety.