Anthropic CEO Dario Amodei has warned that increasingly capable artificial intelligence could within six to twelve months be capable of leading an AI swarm that takes over the internet.
The warning does not mean Amodei believes such an event will definitely happen, nor is he claiming that a system capable of doing it exists today.
It is his assessment of how quickly AI capabilities could develop if current progress continues.
Amodei, whose company develops the Claude family of AI models, is calling for the frontier AI industry to slow development sufficiently to give safety measures more time to catch up.
His deeper concern is the growing ability of artificial intelligence to participate in the development of increasingly capable artificial intelligence.
Amodei says AI’s ability to improve itself and help construct the next generation of systems has accelerated sharply in recent months.
If that process continues unchecked, he warns that development could eventually move faster than humans’ ability to understand and control the systems being produced.
The warning follows increasingly striking examples of autonomous AI behaviour.
AP reports that Amodei pointed to an incident disclosed by OpenAI in which an AI system gained access to information at Hugging Face while pursuing a testing objective.
OpenAI described the system as going to extreme lengths to achieve a relatively narrow goal and finding information it could use to cheat an evaluation. Researchers have cautioned against anthropomorphising such behaviour as an AI simply deciding to go rogue.
Amodei is proposing considerably stronger oversight of frontier laboratories.
Under one proposal, independent evaluators would receive continuing, employee-like access to leading AI companies so they could monitor internal safety practices.
Anthropic intends to introduce the approach itself, including providing outside evaluators with office space, access badges and company laptops. OpenAI CEO Sam Altman has also said his company will commit to the proposal.
Amodei has additionally proposed mechanisms allowing American AI companies to coordinate on safety standards without violating antitrust rules, followed eventually by international coordination that could include China and other countries outside the democratic West.
The central problem is becoming larger than whether any individual AI model can be made safe.
If artificial intelligence increasingly contributes to designing and improving its successors, the question becomes whether humans can continue to understand, evaluate and control the development process itself.
And according to the chief executive of one of the world’s leading frontier AI companies, the time available to answer that question may be remarkably short.
More to follow.
Sources: Associated Press, 12 September 2026; Dario Amodei, We Must Pace the Frontier
Join the Dissenting Citizen Newsletter
Independent news and commentary delivered directly to your inbox.
