Amodei's 6-to-12-month cyberattack warning is a deadline for defenders

Anthropic CEO Dario Amodei has warned that AI models with advanced capabilities could be misused to take control of the internet within 6 to 12 months. That is the claim as reported in the brief note we have. We do not have his full remarks, so we cannot say exactly what threat model he was describing.

Two things give the warning weight. First, it comes from the head of a lab building these systems, not from an outside critic. Second, it names a timeframe. That is very different from vague talk about a superintelligence that may arrive someday. It matches what we keep seeing: the near-term danger from agentic AI is attacks and fraud run at machine scale, not a distant rogue AI.

There is a fair caveat. Frontier labs benefit when their own models look extremely powerful, and 'take control of the internet' is a big phrase. On the cyber benchmarks that actually separate models, the ones built on unguided expert-level tasks, leading models are strong but still fail many tasks. Saturated benchmarks tell us nothing either way.

Our reading: treat this as a deadline for defenders, not a prophecy. The same capabilities that find and exploit vulnerabilities can also find and patch them, and defenders can use them at scale. The coming months will show whether governments and companies put AI to work hardening critical infrastructure, or wait for the first major incident. We take the risk seriously and don't think it is destiny. It is a capability that has to be governed, and the warning should be read that way.