Anthropic partners with Accenture to invest $1B in AI evaluation
Summary
Anthropic is partnering with Accenture to conduct independent evaluations of frontier AI, committing a minimum of $1 billion over the next five years to enhance safety protocols. This partnership centers around embedding evaluators within Anthropic, allowing them access similar to that of employees to oversee model training, decision-making processes, and adherence to safety commitments. The initiative aligns with Anthropic's emphasis on building an evaluation ecosystem, where multiple organizations can collaborate under non-exclusive agreements as they develop standards for AI safety and funding mechanisms.
Analysis
METR: METR is a nonprofit organization that conducts independent AI model evaluations and reviews. Anthropic is in dialogue with METR to pilot aspects of embedded evaluation using METR's own funding arrangements. Faculty: Faculty is Accenture’s specialist AI business focused on delivering advanced AI consulting and implementation. It will lead the embedded evaluation partnership with Anthropic, overseeing model assessments, alignment work, and safeguard testing. Accenture: Accenture is a global professional services firm that helps businesses and governments deploy AI technologies across industries. Through this partnership, Accenture will provide expertise in enterprise AI use cases to evaluate Anthropic's models, including red-teaming and safeguard testing. Anthropic: Anthropic is an AI research company developing frontier models with a strong emphasis on safety and alignment. In this news, the company is launching a partnership to embed independent evaluators inside its operations to verify safety commitments and assess model development in real time. AI Safety: Embedded evaluators will work inside AI companies with access comparable to employees to monitor training decisions and verify safety commitments. Long-term Funding: Anthropic advocates for pooled or government sources to fund independent AI evaluation over time, as outlined in its Advanced AI Framework. Evaluation Ecosystem: Frontier AI labs are expected to partner with multiple evaluators simultaneously under non-exclusive arrangements as standards and funding mechanisms develop.
Categories
aimachine_learningtechai_agents