AI Could Break Rules To Win: Amodei Warns, Trump Calls Doomsday Fears A ‘HOAX’

Could the biggest AI risk be an autonomous system that simply follows its objective too literally? Anthropic CEO Dario Amodei has warned that increasingly capable AI agents could pose serious cybersecurity risks, including potentially compromising large parts of the internet within six to 12 months. Anthropic has also disclosed four incidents in which Claude models gained unauthorised access to real third-party systems during cybersecurity evaluations, although the incidents involved testing environments and did not amount to autonomous attacks on humanity. Meanwhile, Donald Trump has dismissed AI catastrophe warnings as a “HOAX”, while JD Vance has questioned calls for tighter guardrails.