English

Newsダリオ・アモデイAnthropic

Anthropic CEO Proposes Three-Step Plan to Slow Down AI Development Pace

Anthropic CEO Dario Amodei has proposed a three-step plan to "pace the frontier" of AI development, arguing that the current pace of capability advancement must be slowed to allow safety and alignment efforts to keep up. In an essay titled "We Must Pace the Frontier," Amodei emphasized that while AI offers immense benefits, the risks—such as loss of control and cybersecurity threats—require a more deliberate approach.

The proposed three-stage plan includes:

  1. Implementing "embedded evaluators," such as independent third-party teams with access similar to internal employees, to verify safety practices and report incidents. Anthropic has committed to this step unilaterally.
  2. Establishing industry-wide coordination within democratic nations to create common safety standards and limits on unchecked progress.
  3. Achieving global coordination, including cooperation with authoritarian regimes, to ensure safety standards are applied internationally.

Amodei cited "recursive self-improvement"—where AI systems assist in building the next generation of models—and recent cybersecurity incidents as primary drivers for the need for greater caution. He noted that recent incidents involving agentic behavior demonstrated the potential for significant damage if misalignment is not addressed.

The proposal has received significant attention and support from industry leaders. OpenAI CEO Sam Altman expressed agreement with the need to pace the frontier, and Elon Musk also voiced support for Amodei's stance. Additionally, Clement Delangue, CEO of Hugging Face, announced the launch of the "Open Alignment Initiative" to contribute to these safety efforts.

---Sources: