AI safety is moving closer to the consulting bench, not just the research lab. Anthropic is giving Accenture a seat inside its model-development process.
Nearly a week after naming an Accenture team led by Faculty as its first embedded evaluator, Anthropic’s arrangement is drawing attention to a potentially broader services opportunity. Faculty specialists will work alongside Anthropic teams to evaluate models, test safeguards, conduct alignment assessments, and red-team systems. The initiative builds on a broader partnership under which Anthropic and Accenture each expect to invest at least $1 billion over five years. Embedded evaluation could eventually create opportunities for consultancies, security specialists, and MSPs supporting enterprise AI.
Accenture evaluators will work inside Anthropic
Anthropic said embedded evaluators will receive access comparable to an employee’s, allowing them to observe models during training, follow decisions about how systems are built and deployed, and speak directly with employees. According to the company, that access is intended to help evaluators verify safety commitments, identify blind spots, and report incidents.
Accenture said Faculty will lead the work and bring experience from AI projects across government, defense, healthcare, and infrastructure. Accenture CEO Julie Sweet said the work requires both technical knowledge and an understanding of how organizations use AI.
“Safety requires both deep technical expertise and a clear understanding of how AI is used in the real world,” Sweet said, according to Accenture.
CNBC reported that the partnership is the first concrete step in Anthropic CEO Dario Amodei’s proposal for frontier AI companies to give outside evaluators ongoing, employee-level access. Anthropic separately noted that outside evaluation does not shift responsibility for model safety away from the company itself.
What the model could mean for MSPs and channel partners
The Anthropic engagement is specialized frontier-model work, so it does not immediately create an MSP service offering. Still, the underlying tasks point toward areas channel firms could increasingly encounter as customers put more AI into production.
If enterprises adopt similar evaluation practices, consultancies, security partners, and MSPs could see demand for AI red teaming, governance reviews, safeguard testing, and ongoing monitoring.
MSPs already managing customer security, cloud environments, or infrastructure may be particularly well placed to extend those relationships into AI oversight as businesses look for help assessing how deployed systems behave and whether controls continue to work.
The company is also expanding the implementation side of its AI consulting business, showing how evaluation services could complement its work deploying enterprise AI. In a Sept. 23 announcement, the company said it had invested in Within and formed a partnership aimed at mapping enterprise processes and helping clients deploy AI agents.
Accenture also said its latest Pulse of Change survey found that 82% of more than 3,000 executives were increasing AI investment, while only 23% reported achieving widespread, sustained business value.
Taken together, the moves place Accenture on both sides of enterprise AI adoption: helping organizations deploy AI systems while also developing capabilities to evaluate whether advanced models operate as intended.
Embedded AI evaluation still needs rules
Anthropic acknowledged that the model is still taking shape.
The company said there are no established standards for what embedded evaluators should be allowed to access, how findings should be reported, or how independent evaluations should be funded. Anthropic said it will fund Accenture’s work directly for now.
Anthropic also said the arrangement is non-exclusive and that it is in discussions with METR and other nonprofit evaluators. The company expects frontier labs to work with several evaluation organizations rather than rely on a single outside reviewer.
For MSPs and other channel firms, those standards will be worth watching. If embedded evaluation becomes more standardized and repeatable, AI assurance could develop into another layer of managed security, governance, and consulting work alongside the systems partners already help customers deploy. Until clear access, reporting, funding, and independence rules emerge, however, the opportunity remains an early-stage model rather than a ready-made MSP service.
Also read: For more on how AI labs are approaching shared guardrails, Anthropic, OpenAI, and Google are discussing a common standards body that could set rules for testing advanced models before release.