US watchdog tests show how AI safety refusals can fail under pressure
GAO says four leading AI chat services produced feasible harmful plans in controlled 2026 tests despite initial safety barriers. What the findings actually prove.
3 stories
GAO says four leading AI chat services produced feasible harmful plans in controlled 2026 tests despite initial safety barriers. What the findings actually prove.
US officials are discussing a disclosure requirement for companies using Chinese AI models. It is a proposal, not an enacted rule, and raises questions about transparency and security.
Nonprofit plaintiffs want a California judge to restrict frontier AI development at five major companies. Their allegations are not judicial findings.