Anthropic partners with Accenture to evaluate AI model safety
What's the story
Anthropic is partnering with Accenture to assess the safety of its AI models. The move comes as part of a broader effort to slow down the development pace of frontier AI technologies, a proposal put forward by Anthropic CEO Dario Amodei. Last week, Amodei outlined a three-step plan for AI development in his essay.
Investment details
Investment in AI safety
As part of this partnership, both companies are expected to invest at least $1 billion each over the next five years. The funds will be used to build capacity in AI safety.
A team of embedded evaluators will work with Anthropic's internal teams and safety partners as part of this collaboration.
Their responsibilities will include evaluating models, conducting alignment assessments, and testing model safeguards.
Innovative strategy
Embedded evaluators will work from within AI companies
Unlike today's external evaluators, embedded evaluators will work from within AI companies.
This gives them employee-like access to observe how models are trained, the decisions that shape their development and deployment, and interact directly with employees.
Anthropic said in a blog post that these embedded evaluators can assess company operations, verify safety commitments, identify blind spots, report incidents and provide a more informed public account of benefits/risks.
Leadership role
Accenture's faculty to lead the partnership with Anthropic
Accenture's specialist AI business, Faculty, which it acquired earlier this year, will lead the partnership with Anthropic.
Julie Sweet, chair and CEO of Accenture, said their dedicated team has deep AI, security and industry expertise to work alongside Anthropic.
She emphasized that safety requires both technical expertise and a clear understanding of real-world AI applications.
Partnership terms
Non-exclusive partnership
The partnership between Anthropic and Accenture is non-exclusive.
Anthropic will directly fund Accenture's work, citing the "importance and urgency of this work" and the absence of a "settled system for funding independent evaluation."
The company said it plans to collaborate with different evaluators under different funding arrangements as there are no pooled or government sources available at present.