A New Chapter in AI Governance

In a significant move for the technology sector, Anthropic has officially announced the partnership with Accenture to bolster the safety and alignment of its frontier AI models. This collaboration represents the first concrete step in a broader strategy proposed by its CEO to refine how artificial intelligence is developed and monitored.

This initiative follows a proposal outlined by Anthropic’s CEO, Dario Amodei, earlier this month. The strategy, which is detailed in a plan proposed by its CEO, calls for companies to provide ongoing, employee-like access to third-party observers who can ensure that safety practices are rigorously maintained throughout the development lifecycle.

Operational Integration and Investment

The operational structure of this partnership is designed to be deeply integrated. According to Anthropic, Accenture’s evaluators will be tasked with watching models as they take shape during the training phase. These external experts will also be empowered to follow the internal decision-making processes that govern how models are built and deployed, as well as engage in direct conversations with staff members to assess alignment with safety standards.

Financial commitment to this initiative is substantial. Both Anthropic and Accenture have pledged to invest at least $1 billion each over the next five years. This capital is intended to support the human and technological infrastructure necessary to maintain this level of external oversight within the fast-moving AI industry.

The Rationale Behind the Choice

Anthropic justified the selection of Accenture based on the firm's extensive track record in helping both private businesses and government agencies deploy complex AI systems across a variety of sectors. By leveraging Accenture's deep understanding of enterprise-level AI usage, Anthropic aims to gain a more objective perspective on its own model development.

The company has acknowledged that the concept of 'embedded evaluators' is relatively novel and lacks standardized frameworks. Currently, there is no set industry consensus regarding the exact scope of such evaluations, the level of access required by auditors, or the standardized procedures for reporting identified concerns. As an industry leader in the category of AI, Anthropic is essentially pioneering these protocols as they go.

Future Expansion of Oversight

While this partnership is a significant cornerstone of their safety strategy, Anthropic has clarified that it is not exclusive. The company intends to work with a diverse range of organizations beyond Accenture to evaluate its frontier models. The company anticipates announcing additional evaluators in the coming weeks.

Furthermore, Accenture is expected to maintain its independent consultancy status, with the intention of working with other AI companies beyond its engagement with Anthropic. For more developments on this ongoing shift in industry safety standards, readers can follow the latest coverage on Engadget.

Sources

• EngadgetAnthropic picks Accenture for third-party AI safety evaluations