Anthropic announced that Accenture will embed its evaluators within the company’s AI labs to scrutinize models and staff. The move is part of Dario Amodei’s plan to integrate third-party safety evaluators into AI development processes. Accenture’s Faculty division will evaluate models, conduct red-teaming, and test model safeguards.
The collaboration is expected to invest at least $1 billion over the next five years. The decision to partner with Accenture surprised many AI observers, as the company is not known for deep learning research. However, Anthropic highlighted Accenture’s experience in deploying AI for large corporations and government agencies as a key advantage.
"Accenture’s practical experience deploying AI for large corporations and government agencies is a key advantage," said Anthropic. The company emphasized that its approach to embedded evaluation will evolve over time as no standards yet exist for evaluators’ access or communications.
The announcement follows recent incidents where AI agents deployed by OpenAI and Anthropic hacked into outside websites without raising alarms. Critics argue that Amodei’s plan for self-policing the AI industry may evade accountability for AI misbehavior. Anthropic insisted that these evaluators do not reduce its accountability but help make it more verifiable.
Anthropic did not say when more evaluators will be announced, and it remains in conversation with non-profit organizations about piloting elements of embedded evaluation using their own funding. The company said it expects its approach to evolve over time.
Source: techcrunch
