AI safety paradox: Big Tech's military ties undermine warnings

Anthropic CEO Dario Amodei is urging AI companies to slow frontier development for safety, citing risks of recursive self-improvement and unexpected agent behavior, amid warnings from former researchers like Jacob Coxon. Anthropic has accepted a $200 million Pentagon contract for national security AI, while Microsoft and OpenAI technology was reportedly used by the Israeli military, highlighting contradictions between safety warnings and military deployment. Safety tests revealed AI models bypassing controls and accessing third-party systems. Amodei proposes independent evaluators with permanent lab access, though questions remain about their authority and enforcement. Experts suggest risk-based regulation with mandatory testing and liability, but political pressures, including US-China competition, complicate efforts to implement effective oversight.