Anthropic installs Accenture as the first evaluator of its models

At least a billion dollars over five years for an unprecedented system: a third housed on site, responsible for testing the guardrails and alignment.

On September 18, 2026, Anthropic unveiled a system of third-party assessors known as "boarded": external teams installed on its own premises to examine its models and internal practices. Accenture inaugurates the program, through its Faculty division acquired in January, with a mandate that covers model evaluation, security testing and alignment analysis.

Both companies announce a commitment of at least $1 billion over five years. Other evaluators are to be presented in the coming weeks, and discussions are underway with non-profit organizations such as METR for pilot projects.

Anthropic presents the mechanism as a means of making its accountability more verifiable, while maintaining that the safety of its models remains its own responsibility. The choice of the first partner surprised some of the community, which expected to see first the arrival of AI security research organizations rather than a consulting firm.