AI firm Anthropic has partnered with Accenture, embedding its employees to evaluate AI model safety as part of CEO Dario Amodei’s plan to slow AI development. This move addresses growing concerns about AI’s potential risks and comes as Anthropic prepares for a major IPO.
The collaboration is the first step in Amodei’s three-part proposal, aiming to enhance transparency and accountability in AI development amidst increasing industry scrutiny.
Anthropic Taps Accenture for AI Safety Oversight Amid 'Slowdown' Push
In a significant move toward implementing its proposed AI development slowdown, artificial intelligence company Anthropic announced Friday it has selected Accenture as its first embedded evaluator. This strategic partnership will see Accenture employees integrated within Anthropic to rigorously test the safety and ethical alignment of Anthropic's advanced AI models.
The initiative is a direct response to Anthropic CEO Dario Amodei's recently published three-step plan aimed at tempering the rapid advancement of artificial intelligence. The embedding of third-party evaluators marks the first concrete step in this plan, which seeks to address growing concerns about the potential for AI to cause catastrophic harm to humanity. Both Anthropic and OpenAI have faced intense scrutiny from researchers and the public alike regarding AI safety.
While Anthropic and Accenture have committed to investing at least $1 billion each over the next five years in building capacity in this area, Anthropic stated it will directly fund Accenture's work due to the "importance and urgency" of the matter. The company noted that long-term funding should ideally come from pooled or government sources, as outlined in their Advanced AI Framework, but such mechanisms are not yet in place.
Amodei's proposal, unveiled on Saturday, has garnered mixed reactions from industry leaders. While figures like OpenAI CEO Sam Altman and Tesla/SpaceX CEO Elon Musk have shown support, others, such as Nvidia CEO Jensen Huang, have downplayed the need for new regulations. The move comes as Anthropic prepares for an anticipated blockbuster Initial Public Offering (IPO).
The first step of Amodei's plan, which Anthropic has "unilaterally" committed to, involves granting third-party evaluators employee-level access to the company to scrutinize safety practices and report incidents. Anthropic has initially agreed to embed employees from Faculty, Accenture's specialized AI unit, to test safeguards, conduct red-teaming exercises, and assess model behavior against human values. The partnership is not exclusive, with Anthropic in discussions with other third parties, including the research nonprofit METR.
Anthropic emphasized that it retains full responsibility for the safety of its models and that the involvement of embedded evaluators will not diminish its accountability. The company aims to share its process openly to encourage transparency and learning within the AI development community, with plans to evolve its approach as the field matures.
WATCH: Writing the AI safety playbook
The Exchange
Disclaimer: This article contains information from various sources and is presented for informational purposes. Investment decisions should be based on thorough research and professional advice.
