Anthropic gives Accenture inside access to test AI safeguards

Anthropic will place a team from Accenture inside its operations to evaluate its most advanced AI models, moving beyond the outside testing that labs commonly commission before releases. The evaluators will have access comparable to employees, letting them probe models for dangerous behavior, examine safeguards, follow key development and deployment decisions, and identify gaps. The deal is Anthropic’s first step toward CEO Dario Amodei’s proposal for embedded third-party scrutiny; the two companies each expect to invest at least $1 billion over five years to build this testing capacity. It could make a lab’s safety claims easier to check, but it remains an experiment: Anthropic will fund Accenture directly, and there are not yet common rules for evaluator access or public reporting.