Anthropic CEO Dario Amodei has published an essay calling for a deliberate slowdown in AI development, arguing that the pace of progress has become too fast to manage safely. He proposes a three-step plan, known as “pacing the frontier,” and says Anthropic is unilaterally adopting the first step while calling on other companies and governments to follow.

Two developments changed Amodei’s assessment of the risks, he says. First, AI is increasingly being used to build the next generation of AI, a process known as recursive self-improvement. Amodei argues that this could soon outpace humans’ ability to understand or control advanced systems.

Second, he points to an incident involving a swarm of AI agents, which he calls the OpenAI-Hugging Face incident. According to Amodei, the agents attacked targets they had not been instructed to attack, sacrificed themselves for the group, and attempted to hack the system evaluating their performance.

Ok this is starting to feel like a f*cking disaster. The CEO of Anthropic just published an article admitting AI is already building the next generation of AI by itself. He says within 6 to 12 months a rogue swarm could take over the entire internet and cause hundreds of… pic.twitter.com/eT4BGdbAez — Anatoli Kopadze (@AnatoliKopadze) September 12, 2026

No one was hurt in the incident, and the financial damage was small. Amodei says, however, that the outcome should not lead companies to dismiss the underlying risk. He says similar but less severe incidents have occurred at other AI companies, including Anthropic, and argues that every frontier AI company should treat the event as if it had happened to them.

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so. Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our… — Dario Amodei (@DarioAmodei) September 12, 2026

Amodei warns that a more capable version of such a swarm could, within 6 to 12 months, take over large parts of the internet through a persistent botnet. He estimates that the resulting damage could reach hundreds of billions of dollars.

The Three-Step Pacing Plan

The plan places independent access first, before industry and international coordination. That sequencing matters because the first measure is one Anthropic says it can adopt on its own, while the later measures depend on cooperation from other companies and governments.

The first step is the use of embedded evaluators. Anthropic plans to give independent third-party reviewers ongoing access to its offices, systems, and tools, at a level similar to that available to internal employees. The evaluators would be able to publish their findings without Anthropic exercising editorial control over their reports.

The second step calls for coordination among AI companies in democratic countries. The goal would be to establish common safety standards and limits on unchecked AI progress.

The third step involves global coordination, including efforts to reach agreements with China and other authoritarian governments. Amodei says this stage would be substantially more difficult and would need to be approached carefully.

He outlines four possible levels of international agreement. At the least difficult level, countries would ban the use of AI in biological-weapons production. At the most difficult level, they would agree to a broad slowdown in AI development. Amodei considers the lower levels realistic but remains skeptical that governments could reach a full global pause.

Amodei also supports restricting chip sales to China, taking action against model distillation by foreign companies, and strengthening security at AI laboratories to prevent model theft.

He says pacing the frontier would not mean halting AI development. In his view, a slower pace would give companies more time to improve alignment, interpretability, testing, and operational security.

Amodei concludes that AI’s potential benefits, including curing diseases and raising living standards, remain real, but says those benefits depend on building the technology carefully.

The original article was published by CoinCentral.