OpenAI and Anthropic to embed outside auditors to check AI safety

OpenAI and Anthropic will bring in "embedded evaluators" — outside experts with employee-level access to model training. Anthropic named Accenture as its first evaluator; each company pledges at least $1 billion for AI safety over five years. The oversight remains voluntary.

• Evaluators get access to training and staff
• Anthropic named Accenture its first embedded evaluator
• Each company to invest at least $1B in safety over 5 years
• Oversight is voluntary, with labs setting its scope