Anthropic has partnered with Accenture to establish an independent evaluation programme for frontier artificial intelligence (AI) models, with a focus on model safety, alignment and safeguards. The initiative will be led by Faculty, Accenture’s specialist AI business, and forms part of Anthropic’s efforts to introduce independent evaluators within the company.
Under the partnership, evaluators will assess and red-team AI models, conduct alignment assessments and test model safeguards. Accenture’s experience in deploying AI across businesses and governments is expected to bring an enterprise perspective to the evaluation of how frontier models are developed and used.
Anthropic and Accenture each expect to invest at least $1 billion over the next five years to build capacity for independent AI evaluation.
The initiative introduces an embedded evaluation model in which independent evaluators work within AI companies with access comparable to that of employees. According to Anthropic, this could allow evaluators to observe models during training, examine decisions related to their development and deployment, and engage directly with employees.
Such access is intended to enable evaluators to assess how AI companies operate, verify safety commitments and identify potential blind spots. They could also report incidents and provide the public with additional information about the potential benefits and risks associated with frontier AI systems.
Anthropic noted that embedded evaluation remains an emerging practice, with no established standards governing the information evaluators should access or how findings should be reported. The company also said there is currently no settled model for funding independent evaluation.
For the initial programme, Anthropic will directly fund Accenture’s work. The company is also in discussions with nonprofit evaluators, including METR, to pilot aspects of embedded evaluation through independent funding arrangements.
Anthropic said the partnership is non-exclusive and that it plans to work with additional evaluators. Accenture will similarly work with other AI developers. The company said its longer-term objective is to support an ecosystem of independent evaluators operating under shared standards as frontier AI development advances.
