Anthropic has selected Accenture as its first embedded evaluator, granting the consulting firm's employees access inside the AI lab to scrutinize models and safety practices as part of a commitment worth at least $1 billion over five years1,2.
The arrangement is the first concrete step toward the third-party oversight framework that Anthropic CEO Dario Amodei laid out in a recent proposal to slow the pace of AI development. Under the plan's first step, outside evaluators receive employee-level access to verify safety practices and report incidents.
Who does the work
Faculty, the specialist AI business Accenture acquired in January, will carry out the hands-on evaluation. Its staff will red-team models, conduct alignment assessments, and test model safeguards from inside Anthropic. Anthropic said it will fund Accenture's work directly, noting that pooled or government funding sources it has called for do not yet exist.
The partnership is not exclusive. Anthropic said more evaluators will follow in the weeks ahead and that it is in discussions with the research nonprofit METR and other third-party organizations about piloting elements of embedded evaluation using their own funding.
Why Accenture, not a safety lab
The choice drew attention because the public discussion around embedded evaluators had centered on AI safety research organizations such as METR, Redwood Research, and Apollo Research. Anthropic pointed to Accenture's practical experience deploying AI for large corporations and government agencies as a key advantage, and noted that Accenture, as a large public company that predates the current AI wave, offers a degree of functional independence.
Accenture's shares rose 8% in after-hours trading following the announcement.
Safety scrutiny in context
The embedded-evaluator program arrives as leading AI labs face mounting pressure over model safety. OpenAI on September 16 published six previously unreported incidents of model misalignment observed during training and evaluation over the past six months, alongside a new disclosure framework for such cases[3]. ANALYSIS Anthropic's move to formalize outside access for safety auditing represents a structurally different approach from OpenAI's self-reporting framework, placing third-party personnel inside the lab rather than relying on internal disclosure.
Anthropic described the initiative as part of the commitment outlined in its CEO's essay on slowing AI development, with embedded evaluation serving as the first of a three-step plan3.