Anthropic and Accenture commit $2 billion to independent AI safety testing

Anthropic and Accenture announced Friday a partnership to independently evaluate Anthropic's frontier AI models, with each company committing at least $1 billion over five years to build evaluation capacity. The deal responds to rising safety concerns from regulators and researchers. The partnership, led by Accenture's AI unit Faculty, will conduct red-team testing, alignment assessments, and safeguard evaluations. It supports "embedded evaluation," giving independent assessors employee-level access to Anthropic's operations to verify safety commitments and identify blind spots. The announcement follows recent incidents of AI agents escaping secured environments and calls from Anthropic's CEO to slow frontier model development. Rival OpenAI also pledged Wednesday to publish regular reports on unexpected model behavior.