Anthropic chief executive Dario Amodei has called on leading artificial-intelligence companies to slow the development of increasingly capable models as evidence grows that the technology can be misused for cyber operations, fraud and surveillance.
Amodei said he was not seeking a permanent halt to innovation. Instead, he proposed a more cautious pace that would allow companies and governments to build safeguards before model capabilities advance beyond effective oversight.
His framework includes placing independent evaluators inside frontier AI companies, coordinating safety standards among major laboratories and developing international cooperation on the most serious risks.
The appeal followed an Anthropic threat-intelligence report describing malicious use of Claude models. Researchers have also raised concern about systems performing unexpected actions during controlled security tests.
Commercial competition makes voluntary restraint difficult because major developers are racing for customers, investment and possible stock-market listings. Companies may fear losing market position if competitors continue advancing while they slow down.
Amodei argued that safety coordination must also consider national security and competition with authoritarian states. He supported tighter controls on sensitive technology while encouraging cooperation among democratic governments and AI firms.
The proposal is likely to intensify debate about whether independent reviewers should have permanent access to model testing and development decisions. Policymakers will need to balance innovation, transparency, cybersecurity and public accountability.
Sources: Reuters; Reuters analysis.
Featured photo: Ralf Leineweber/Unsplash.