Particle.news

Anthropic Embeds Accenture Evaluators Inside Its AI Lab

The company says the pilot will give outside experts employee-level access to training and safeguards to make its safety claims easier to verify.

Overview

  • Anthropic announced a pilot with Accenture’s Faculty that will place external evaluators inside its lab to run red teams, alignment checks and safeguard tests.
  • The partners said they expect to invest at least $1 billion over five years and Anthropic will fund the initial evaluation work directly.
  • Faculty staff will have access comparable to Anthropic employees, including the ability to observe model training and review development and deployment decisions.
  • The arrangement is non-exclusive and Anthropic said it is talking with nonprofit evaluators such as METR and plans more partner announcements.
  • Critics warn company-funded embedded teams could look like self-policing and leave open questions about evaluator independence, reporting standards and long-term governance even as markets responded positively.