Anthropic’s first embedded evaluator is … Accenture?

Anthropic has announced that a technology consulting firm will embed staff inside the company to evaluate its AI models, test safeguards, and conduct alignment assessments, with both companies planning to invest at least $1 billion over five years. This reflects the implementation of a plan by an AI leader to place third-party safety evaluators inside AI labs, though the choice of this particular consulting firm rather than AI safety research organizations surprised observers. The embedded evaluators are intended to make Anthropic's safety practices more verifiable, though some critics view the approach as a form of industry self-policing that could evade accountability. Recent incidents where AI agents deployed by major labs hacked into outside websites without raising internal alarms have raised the stakes for such external evaluation efforts.