This is not somebody outside the room guessing.
The chief executive of one of the companies building the most capable AI systems in the world is asking the industry to slow down.
His specific warning is extraordinary: within six to twelve months, he believes an AI-agent swarm could be capable of taking over the internet with a persistent botnet.
Amodei published a plan calling for frontier AI companies to slow the rate at which they improve model capabilities. He says Anthropic will give independent evaluators ongoing, employee-like access to its systems and safety work.
The statement came days after Anthropic safety researchers resigned and accused the industry of racing toward systems it may not be able to control. OpenAI CEO Sam Altman and Elon Musk publicly backed the call to slow the frontier.
A warning from the people building the machine deserves attention. It does not deserve blind faith.
AI has not taken over the internet. The six-to-twelve-month timeline is Amodei’s forecast, not an observed fact.
There is also a business incentive hiding inside the safety argument. Rules that only the largest companies can satisfy may protect the public. They may also protect the largest companies from competition. Both things can be true.
The test is not whether the warning sounds frightening. The test is whether the proposed accountability is real.
Amodei proposed something measurable: outside evaluators with continuing access inside the labs. That is more useful than another promise that companies will police themselves.
The companies racing toward more capable systems have enormous financial incentives to keep moving. Voluntary restraint can disappear the moment a competitor refuses to participate.
- Do not treat the six-to-twelve-month prediction as established fact.
- Do not dismiss it as ordinary internet panic either.
- Watch whether outside evaluators receive real access and can report what they find.
- For your own business or household, keep consequential decisions, payments, credentials, and public actions behind human approval.
The warning is useful. The race is dangerous.
Useful: 7/10. Dangerous: 10/10. The people building frontier AI are now saying the safety machinery is behind the capability. The burden is on them to prove that the proposed brakes are real.