Archive · 12 Sep 2026
Anthropic CEO Urges Slower AI Development Amid Misuse Fears
Dario Amodei leads Anthropic, a company that builds artificial intelligence models.
He says AI companies should improve their models more slowly.
He does not want companies to stop training models completely.
Instead, he wants them to spend more time making the models safer.
He also wants independent experts to check how companies measure and manage risks.
Anthropic said some people had used its Claude models for harmful activities.
These activities included cyber operations, surveillance, fraud and weapons development.
Other AI leaders and researchers have also expressed concern about possible dangers.
Amodei wants AI companies to cooperate on safety rules while governments consider new regulations.
Anthropic CEO Dario Amodei urged AI companies to slow capability improvements rather than halt development.
Amodei proposed more time for model alignment, safeguards and independent evaluation.
Anthropic reported Claude models had been used in weapons development, cyber operations, surveillance and fraud.
The call followed warnings from Anthropic researcher Jacob Coxon and similar openness from OpenAI CEO Sam Altman.
Amodei proposed permanent third-party reviewers inside frontier AI companies and voluntary industry standards.
- Who
- Dario Amodei, Anthropic's CEO, along with AI companies, independent reviewers, researchers and lawmakers.
- What
- Amodei called for a slower pace of AI model development and proposed a three-step framework for managing risks.
- When
- The call was made in an essay shared on X after Anthropic released a threat intelligence report on Thursday; related concerns emerged this week.
- Where
- The proposal concerns frontier AI companies and comes amid debate among U.S. lawmakers.
- Why
- Because AI models have reportedly been used for harmful activities and their growing capabilities may be difficult to contain.
Slow Development and Expand Oversight
Continue Progress with Safeguards
Pace of AI advancement
Slow Development and Expand Oversight
Amodei argues that companies should slow improvements in model capabilities to create time to manage risks.
Continue Progress with Safeguards
Amodei clarified that he is not calling for a halt to model training or technical progress, but for continued development with adequate safety measures.
Who should monitor risks
Slow Development and Expand Oversight
Permanent third-party reviewers should have access to relevant tools and internal risk-assessment processes at frontier AI companies.
Continue Progress with Safeguards
AI companies should continue leading technical development while voluntarily cooperating on shared standards and safeguards.
- Speaker
- Dario Amodei, CEO of Anthropic
- Main proposal
- Slow the pace of AI capability improvements without halting model training
- Safety measures
- Allow more time for alignment, safeguards and third-party evaluation
- Reported misuse
- Anthropic said Claude models were used for weapons development, cyber operations, surveillance and fraud
- Independent oversight
- Amodei proposed permanent third-party reviewers inside frontier AI companies
- Industry response
- OpenAI CEO Sam Altman said his company was open to slowing development
- Policy response
- Amodei urged voluntary industry standards as U.S. lawmakers consider AI rules