Jane Edwards 
- Anthropic and Accenture have partnered to launch embedded AI evaluation
- Faculty will handle various testing processes, including model assessments
- The 2026 FedCiv Summit will cover AI modernization and information sharing
Anthropic and Accenture have partnered to perform embedded evaluation of Anthropic’s frontier artificial intelligence systems, with Accenture’s Faculty business unit leading efforts to evaluate, red-team and assess the safety of the models.
As federal agencies weigh how to responsibly adopt increasingly capable AI tools, discussions around oversight, safety and mission integration continue to shape federal technology strategy. The Potomac Officers Club’s 2026 FedCiv Summit on Oct. 29 will bring together government and industry leaders for a series of conversations on AI-driven modernization, mission delivery and interoperable, trusted information sharing across federal civilian agencies. Register now to join the conversation shaping the future of federal civilian technology.
Anthropic said Friday the initiative builds on a commitment Anthropic CEO Dario Amodei outlined in his essay to embed evaluators within the company.
What Is Embedded Evaluation?
Embedded evaluation differs from external review in that evaluators work within the AI company with employee-like access rather than reviewing models from outside after the fact. Evaluators are able to observe models as they undergo training and track the decision-making behind model development and deployment. Evaluators can also communicate with employees.
Independent embedded evaluators are meant to ensure that accountability is confirmable. However, the process is new, so various details, including information access, reporting and funding, have yet to be finalized.
What Does the Partnership Involve?
Under the partnership, Accenture’s Faculty unit will lead independent assessment of Anthropic’s AI models, including stress-testing, alignment checks and safeguard testing.
Both companies plan to invest at least $1 billion over five years to grow their capacity for this kind of work. Anthropic will cover Accenture’s costs itself, citing the urgency of the effort and the lack of an established funding model. Anthropic is also talking with METR and other nonprofit groups about similar, self-funded pilots.
The partnership is not exclusive.
How Does the Partnership Align With Accenture & Anthropic’s Broader AI Work?
The partnership builds on a relationship the two companies formed in December, when Accenture and Anthropic established the Accenture Anthropic Business Group, a joint entity aimed at accelerating the transition of AI from pilot to full-scale deployment and developing new offerings for regulated industries, including the public sector. As part of that effort, the group set out to train approximately 30,000 Accenture staff on Claude.
Since then, each company has continued building out its own AI work separately. Anthropic brought on Teresa Carlson as global head of public sector in July, and has been positioning its most restricted model under Project Glasswing and Claude Mythos for vetted government and industry partners.
Accenture Federal Services named Garrett Berntsen as chief AI officer in January and struck an AI adoption partnership with OpenAI in May. It has also won a potential five-year, $821 million task order to provide core integration support for the Department of War Chief Digital and Artificial Intelligence Office’s War Data Platform, as well as a potential $480 million Army contract to develop the AI-enabled Joint Enterprise Task Management System.
