AI Watchdogs Needed: CEO Calls for External Oversight

Anthropic CEO Dario Amodei believes independent external watchdogs are crucial for safeguarding developing AI technologies. He has proposed a system of "embedded evaluators" who would receive employee-like access within AI development companies. Their role would be to verify adherence to safety practices, report incidents, and assess the alignment of AI models and their development pipelines.

Organizations like METR, an AI safety testing laboratory, are presented as examples of such potential oversight bodies. METR's founder and CEO, Beth Barnes, previously worked at OpenAI on early ChatGPT versions. Paul Christiano, who also led research at OpenAI, founded an earlier iteration of METR. Figures like Barnes and Christiano have informal ties to Anthropic, which received early funding from prominent supporters of the "effective altruism" movement.

Amodei, who left OpenAI in 2020 due to concerns about the company's safety guardrails and transparency, co-founded Anthropic with AI safety as a core principle. The company has previously declined contracts it deemed unsafe, including work for the U.S. Department of Defense. Amodei stresses that while AI holds immense potential to improve human life, this benefit hinges on building the technology responsibly.