Amodei warns of mass cyberattacks within 6 to 12 months: a risk worth governing, a claim that needs details

Dario Amodei, CEO of Anthropic, has warned that AI models with advanced capabilities could be used to take control of the internet within 6 to 12 months, according to the brief report we have. The material does not say what he means by 'taking control', which models he is referring to, or what evidence supports the timeline.

The context gives the warning weight. In cybersecurity benchmarks that are not yet saturated, meaning hard expert-level tasks completed without human guidance, frontier models already score high. In our index, GPT-5.6 has 71 points and Anthropic's Mythos has 69. The risk is not science fiction. As we have argued before, the immediate danger is the automation of fraud and intrusion at industrial scale, not a distant superintelligence.

Our reading: take the warning seriously, but keep a source-critical eye. It comes from the head of a company that is going public and whose identity rests on safety, so the message also serves its positioning. That does not make it false. It does mean defenders, regulators and vendors should ask for specific evidence. 'Six to twelve months' is useful only if it leads to action: patching critical infrastructure, auditing agents and giving defenders the same models attackers have. Over the long run, the same capability that breaks systems can secure them at scale. Whether it does depends on who adopts it first.