Anthropic is putting Accenture inside its AI lab to test models

Faculty will work where the models are built

Anthropic says the partnership will be led by Faculty, Accenture's specialist AI business. Its staff will evaluate and red team models, conduct alignment assessments, test safeguards and speak directly with the employees building and deploying them.

That access is the central change. Traditional outside evaluators usually receive a defined model or a limited testing window. Embedded evaluators will work inside the lab with access comparable to an employee, allowing them to watch systems take shape during training and examine the decisions around deployment.

A new kind of AI safety job

The announcement points to a labor market that is broader than model research alone. Anthropic is buying capacity for people who can combine technical testing with alignment analysis, incident reporting and practical knowledge of how companies use AI in the real world.

Accenture said its experience deploying AI for businesses and government agencies gives Faculty a view that a research nonprofit may not have. For workers, that suggests safety work will increasingly value both deep model knowledge and the ability to test systems in operational settings. The companies did not disclose a headcount or a hiring plan, so the investment figure should not be read as a promise of a specific number of new jobs.

Independence is the unresolved test

Anthropic says it will fund Accenture's work directly because no settled system exists for financing embedded evaluation. The lab also says it is talking with METR and other nonprofit evaluators, and expects to announce more organizations in the coming weeks.

The open question is how an evaluator can work inside a lab without becoming an extension of its communications team. Anthropic acknowledges that there are no common standards for access, reporting or funding yet. It says evaluators will make safety commitments more verifiable, while responsibility for the models remains with Anthropic.

That makes the first phase a live experiment for both companies. The jobs may sit inside a consulting group, but their credibility will depend on whether evaluators can document problems, challenge decisions and report findings with enough independence to be trusted outside the lab.