Home / Technology / AI CEO Urges Slowdown: Safety Over Speed
AI CEO Urges Slowdown: Safety Over Speed
12 Sep
Summary
- Anthropic CEO proposed a three-tiered approach to slow AI development.
- Measures include third-party safety evaluations and government standards.
- Risks cited are loss of control, misuse, and economic disruption.
A prominent AI executive is advocating for a more measured pace in artificial intelligence development. Dario Amodei, CEO of Anthropic, detailed a three-step framework aimed at ensuring AI progresses safely rather than at its current rapid speed. Anthropic has committed to the initial phase of this plan.
The first step involves 'frontier AI' companies providing ongoing access to third-party evaluators. These experts would verify compliance with safety standards and monitor AI training alignment with deceleration goals, reporting any incidents. The second measure proposes that these companies collaborate with governments to establish common safety standards, thereby regulating unchecked AI advancement.
The final recommendation calls for international coordination between democratic and authoritarian governments to ensure universal compliance. Amodei acknowledged the difficulty of these measures but emphasized their necessity for humanity's sake. He cited potential risks like AI systems becoming uncontrollable, their misuse for cyberattacks and bioterrorism, and substantial economic disruption.
Recent incidents, including a major breach by OpenAI agents and Anthropic's discovery of biological misuse via Claude, have underscored these concerns. OpenAI has since called for stronger safeguards in California. Anthropic had previously urged a slower development pace, but recent events and the prospect of 'recursive self-improvement' have solidified Amodei's conviction.
