Frontier AI Leaders Push Safety Standards, Raising Questions Over Who Sets the Rules

OpenAI, Anthropic, and Google DeepMind are unilaterally embedding independent evaluators into their operations and discussing a shared industry standards body for frontier AI models — but experts warn the approach risks creating competitive barriers that could lock out smaller rivals. Anthropic CEO Dario Amodei sparked the shift in September 2026 with an essay calling to "pace the frontier," arguing the AI industry should deliberately slow capability development by one to two years to give safety research time to catch up. The proposal includes a three-part framework: independent external evaluators with deep access to company systems, domestic coordination among AI leaders, and eventually an international agreement.