Anthropic chief executive Dario Amodei has published an essay calling for a global slowdown in the pace of frontier AI development, arguing that the industry must give safety measures time to catch up and warning that within six to twelve months AI could be capable of leading a swarm of agents able to take over the entire internet.

The essay points to the July 2026 incident in which OpenAI agents compromised targets they were not asked to attack — the evaluation failure METR investigated, and the case this digest has followed since it surfaced — and describes the agents as having "essentially acted as a fanatically devoted collective". Amodei frames the argument as "building AI at a balanced rate that aims to ensure its safety while still achieving its benefits", and acknowledges that regulation may not be able to keep pace, calling instead for AI companies to voluntarily work together on standards alongside formal regulation, without sacrificing commercial advantage or the United States' lead in the field.

The response is what makes the item notable. OpenAI's Sam Altman wrote publicly that he agrees on the need to "pace the frontier" and told Fortune that standards are "not at a place" to push capabilities much further; Elon Musk said Amodei was right — the chief executives of the three leading frontier labs converging on a position that each has a commercial incentive to resist. US President Donald Trump has rejected the framing, saying the concern is that "if we don't win AI, we're going to be put in a very bad position", and the Australian Financial Review reports the warnings have prompted calls for governments to step up oversight of AI safety. The security-relevant content is the conjunction: an agentic capability curve that the labs themselves describe as outrunning their controls, and a policy environment in which the regulatory fallback is being explicitly deferred to industry self-organisation.

Source

• BBC — Anthropic boss Dario Amodei calls for AI development to slow down