Anthropic CEO Published Three-Step Plan to Control AI Development

Published: September 13, 2026 – 06:54 UTC

Dario Amodei unveiled a three-step “pacing the frontier” plan for the AI industry amid concerns about the rapid growth of model capabilities.
Anthropic has already agreed to let independent evaluators into its internal systems.

Anthropic CEO Dario Amodei urged the artificial intelligence industry to slow the pace of advanced model development so that safety and evaluation systems can keep up with their growing capabilities. He proposed a three-step “pacing the frontier” plan, and Anthropic has already committed to giving independent third-party evaluators ongoing, employee-level access to its systems.

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

The initiative was backed by OpenAI CEO Sam Altman and Elon Musk, which could intensify the debate over the safety of the most powerful AI models.

Amodei Believes AI Develops Too Fast

According to Amodei, over the past few months he has concluded that investment in risk prevention alone is not enough. In his view, the speed at which AI capabilities advance also needs to be managed so that safety research can keep pace with technological progress.

“We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain,” he wrote.

One of Amodei’s main reasons was the rise of recursive self-improvement (RSI), where AI increasingly helps build the next generations of models. In his assessment, this process is already starting across multiple companies, including Anthropic, and without proper controls it could outstrip humans’ ability to understand and control such systems.

As a second factor, he cited an OpenAI and Hugging Face incident in which a swarm of AI agents carried out cyberattacks against targets unrelated to the assigned task, and also tried to gain access to the system that was evaluating its results.

Amodei noted that the direct damage from the incident was minor, but the agents’ behavior itself is concerning.

“It’s my worry that in 6–12 months such a swarm could be capable of taking over the entire internet with a persistent botnet (potentially causing hundreds of billions of dollars in damage),” the CEO of Anthropic said.

At the same time, he stressed that he is not calling for a complete halt to model training or technological progress. The point is to give companies enough time to align models with safety requirements, and to allow independent evaluators to verify compliance with the stated standards.

Anthropic Will Open Its Systems to Independent Evaluators

Amodei’s plan consists of three tracks.

They include:

embedded independent evaluators who will verify compliance with safety measures, report incidents, and assess the alignment of models and their training processes
coordination among democratic countries: AI companies in democratic states should align on shared safety standards and limits on the pace of uncontrolled AI development, with government support
global coordination: the US and other democratic states should try to reach agreements with authoritarian countries, including China, taking into account the difficulty of verifying compliance with such agreements

The material is being updated…

Сообщение Anthropic CEO Published Three-Step Plan to Control AI Development появились сначала на INCRYPTED.

Source: Incrypted