From CNBC · Ashley Capoot · · 1m 30s
Anthropic taps Accenture to police its own AI safety
The AI safety company embeds outside evaluators to test models, implementing CEO Dario Amodei's plan to slow development.
In brief
The AI safety company embeds outside evaluators to test models, implementing CEO Dario Amodei's plan to slow development. Anthropic is putting independent eyes inside the company to verify safety practices, a concrete first step toward slowing AI development that other companies may follow. Originally reported by CNBC.
Anthropic picks Accenture as safety watchdog
Anthropic announced it will embed Accenture employees inside the company to test AI model safety, red-team systems, and verify they align with human values.
Amodei's three-step slowdown plan
Anthropic CEO Dario Amodei published a proposal to temper how quickly AI companies improve their most advanced models. Embedded evaluators are the first concrete step.
Industry splits on safety concerns
OpenAI CEO Sam Altman and Elon Musk backed Amodei's proposal, but Nvidia CEO Jensen Huang dismissed concerns and argued against new regulation.
Anthropic and Accenture commit $1 billion
Both companies agreed to invest at least one billion dollars over five years to build evaluator capacity. Anthropic will fund Accenture's work directly for now.
Anthropic remains accountable for safety
The company emphasized it is still fully responsible for model safety and that embedded evaluators will not reduce its accountability. More evaluators are being discussed.
“We're sharing these early efforts now so people and other AI developers can see our process.”
Long-term funding from government sources
Anthropic called for pooled or government funding for evaluators in its Advanced AI Framework in June. It plans to work with different evaluators under different funding arrangements.
The takeaway
Anthropic is putting independent eyes inside the company to verify safety practices, a concrete first step toward slowing AI development that other companies may follow.
