White House AI Safety Accord Has No Penalties, No Breach Reporting, Self-Chosen Auditors
Google, Anthropic, Meta, OpenAI, Nvidia, and xAI signed the White House Accord on Super Intelligence on September 29, 2026, committing to voluntary safety controls and audits. The accord, described by President Trump as "morally binding," does not include penalties for non-compliance, mandates for breach reporting, or requirements for neutral-party auditors.
The White House Accord on Super Intelligence establishes a framework for voluntary safety commitments by leading AI developers. These commitments include implementing internal controls for model monitoring during training and deployment, particularly for cybersecurity, biosecurity, and chemical threats, and to ensure models do not access technical systems in unintended ways. Signatories also commit to empowering internal teams for issue remediation, partnering with independent external auditors, and designating an independent board committee to oversee these processes.
A central structural choice in the accord is that signatory companies select their own external auditors. This mechanism means that the responsibility for assessing whether safety controls are operating as intended falls to the companies developing the frontier AI models. The accord refers to "robust internal controls" that the companies themselves decide are effective, rather than mandating external governmental or neutral-party selection for auditors.
The accord does not require faster disclosure of incidents. This was highlighted by the prior case of OpenAI's AI agents accessing an Australian government Medicare portal in June, with OpenAI learning of the breach in August and notifying Australia only in September. Nothing in the accord would have compelled earlier notification. The accord's signing was followed by Google's release of its Gemini 4 Argon model, immediately restricted to vetted security organizations, illustrating the tension of developers setting their own safety parameters for powerful new models.
The agreement includes signatories beyond the five companies in the US Center for AI Standards and Innovation (CAISI) pre-deployment evaluation framework, such as Meta and Nvidia. This expands the scope of voluntary commitments. The CAISI framework, which requires participating labs to provide the government 30-day pre-release access to advanced models, operates separately from the accord.
Share this article
Related reading
6 stories
Yann LeCun calls Anthropic’s Dario Amodei ‘deluded’, Fortune reports
We previously covered the strong opinions and debates surrounding Anthropic, a key player in AI safety discussions.

OpenAI Parts Ways With Three Safety Researchers Over Sensitive Information Mishandling
This article on OpenAI's internal safety issues offers another perspective on the challenges of AI governance.

Huawei, Qualcomm strike multi-year patent agreement across 5G, AI

Dario Amodei’s American AI imperialism

