Anthropic has reportedly taken a first step toward the safety framework its chief executive outlined earlier this month, naming Accenture (NYSE: ACN) as an on-site independent reviewer of its most advanced systems.
The arrangement is meant to put outside scrutiny inside the lab itself, rather than leaving assessment to periodic external audits after models are already trained.
The move follows Dario Amodei’s September 12 essay arguing that frontier laboratories should deliberately moderate the speed of capability gains so that oversight can keep pace.
The first element of that plan was “embedded evaluators”: third-party teams granted access comparable to employees so they can observe training runs as they happen, track the decisions that shape how models are built and released, and talk directly with staff.
Anthropic said it was already prepared to adopt that practice on its own.
The Accenture partnership is the first public implementation.Faculty, the applied-AI unit Accenture acquired earlier this year, will lead the work.
Its staff are expected to evaluate and red-team models, run alignment assessments, and test existing safeguards.
Anthropic has emphasized Faculty’s experience helping large organizations and governments put AI into production, arguing that practical knowledge of how systems are actually used will strengthen the safety lens applied to its own research.
Both companies say they expect to commit at least one billion dollars each over five years to build the necessary capacity.
Because no established pool of public or industry funding for this kind of independent review yet exists, Anthropic will pay for Accenture’s work directly in the near term.
Officials described the arrangement as non-exclusive.
Conversations continue with nonprofit groups such as METR about self-funded pilots, and additional evaluator partners are expected to be named in the coming weeks.
Accenture, for its part, will be free to perform similar roles for other developers.
The operational details remain unfinished.
Embedded evaluation is a new practice, and questions about access rules, reporting channels, and what findings can be made public have not been fully settled.
Anthropic’s announcement presents the partnership as an early experiment rather than a finished standard, one that should eventually sit inside a broader ecosystem of evaluators working to common criteria.Industry reaction has been mixed.
Some observers noted that safety researchers had expected the first embedded team to come from a specialized nonprofit rather than a large consultancy that already has commercial ties to Anthropic.
Supporters of the choice point to Accenture’s scale, its independence as a long-established public company, and Faculty’s prior safety work with leading labs.
Market reaction was immediate: Accenture shares rose sharply in after-hours trading after the news.
The announcement does not itself slow training schedules.
It is framed as infrastructure that would make any future slowdown more credible by giving outsiders a continuous view of how safety commitments are kept. Whether other frontier labs follow with comparable access remains an open question.
For now, Anthropic has converted a policy essay into a funded, on-site program and invited others to help define the rules as the work proceeds.