THE leaders of six major AI companies—OpenAI, Google, Meta, Nvidia, xAI and Anthropic—have signed a voluntary agreement with President Trump to strengthen safety measures around frontier AI systems. The White House Accord on Super Intelligence asks these firms to establish internal safety controls, subject those controls to independent audits, and give boards greater oversight of AI safety and security.
The framework outlined is non-mandatory and does not set specific implementation deadlines, though officials suggest that over time such steps could be codified into law or regulation.
The accord proposes a four-layer oversight model for frontier labs: first, internal controls to monitor a model’s capabilities during training and deployment with attention to cybersecurity, biosecurity and chemical threats; second, a dedicated internal team to ensure controls are functioning and issues are fixed; third, an independent external auditor to verify internal controls; and fourth, an independent board committee to oversee the process and ensure remediation.
Experts quoted in the piece describe the plan as a positive, but limited, step, emphasising that voluntary compliance and the absence of enforceable deadlines or public reporting reduce its impact. They note that meaningful oversight requires genuinely independent evaluation and enforceable remediation of identified problems, rather than “security theatre.” The article notes the outcome hinges on how the companies implement and demonstrate real accountability in practice.