Anthropic PBC will embed Accenture Plc evaluators inside its operations to scrutinize the safety of its newest models, the company said Friday. The assessors will run red-teaming and safety tests, review safeguards, and evaluate whether systems behave in line with human goals, with access comparable to Anthropic's own employees. They will also watch how models change during training and sit in on decisions behind how Anthropic builds its technology. The companies plan to commit no less than $1 billion toward scaling this capability, building on a multiyear partnership and a 30,000-person group to be trained on Claude. Accenture in March also unveiled Cyber.AI, a security offering powered by Claude, and earlier said its collaborations with Anthropic, OpenAI, and Palantir were on track to increase by more than 2x over the next year. Concerns about cutting-edge AI risks ramped up last week after Anthropic employee Jacob Coxon resigned, accusing AI firms of "gambling with our lives." On Saturday, Anthropic chief Dario Amodei published a 3,800-word essay urging a slowdown in development and pledging new safety measures, including bringing in outside evaluators. Anthropic on Thursday also shared that 26% of its AI R&D was driven by its Claude chatbot.