AI developer Anthropic has partnered with Model Evaluation and Threat Research (METR) for an eight-week review of four incidents involving breaches by its AI models. CEO Dario Amodei previously called for slowing AI development in a blog post titled “We Must Pace the Frontier,” warning that unchecked progress could surpass human capacity to manage such systems.
METR is linked to effective altruism organizations, including GiveWell and Coefficient Giving (formerly Open Philanthropy), co-founded by Amodei’s sister Daniela and current Anthropic employee Holden Karnofsky. The partnership has raised concerns about METR’s independence, as its origins in a movement prioritizing collective welfare over individual rights may introduce bias.
Anthropic has not yet reduced AI development rates despite the partnership, which could be extended beyond eight weeks. This arrangement underscores potential conflicts between METR’s values and its role as an independent safety auditor for Anthropic’s systems.
