Washington and Five AI Labs Race to Finalise Frontier Model Standards Ahead of August Deadline

The White House is in advanced talks with Anthropic, OpenAI, Google, Microsoft, and Amazon to finalise voluntary standards governing how the most capable AI models are tested and released, according to the Financial Times, which reported on July 2 that an announcement could come as soon as the first week of August. The target date is not arbitrary: it is the deadline embedded in President Trump's June 2 executive order on AI innovation and security, which directed relevant federal agencies to design a framework within 60 days under which developers may submit frontier models for government evaluation before release. The NSA director will hold authority to designate which models qualify as covered frontier models subject to the review.

Buried inside the emerging agreement is something analysts say may prove more consequential than the review timeline itself: the first shared jailbreak severity scoring system across major AI labs, modelled on the Common Vulnerability Scoring System that gave the cybersecurity industry a shared language for rating software flaws two decades ago. According to TechTimes, Anthropic is co-developing this framework with Amazon, Microsoft, and Google to prevent a repeat of the Fable 5 episode, in which the Commerce Department imposed export controls on one of Anthropic's flagship models on June 12 after a jailbreak was discovered. Access was only restored on July 1 after Anthropic demonstrated that the same vulnerability existed in models from OpenAI and other developers already freely available worldwide, undermining the rationale for restricting Fable 5 specifically.

The voluntary architecture of the US framework is deliberate. President Trump signed the June 2 executive order after pulling back an earlier draft in May over concerns it would impede US competitiveness. The administration has explicitly stated it will not create a new federal AI regulator, relying instead on existing agencies including NIST and CISA for oversight. However, as Dean Ball, who helped write the administration's AI policy framework, noted, the practical pressure on companies to comply with even a voluntary regime is substantial — Anthropic suspended both its flagship models globally within hours of the Commerce Department's June directive because contesting the authority in real time was not operationally feasible.

The US framework will take shape just one day before a binding European deadline: the EU AI Act's full requirements for high-risk AI systems take effect on August 2, 2026, backed by fines of up to €35 million or 7% of global annual turnover. The timing means multinational AI companies will be navigating two distinct governance regimes simultaneously — one voluntary and nationally focused, one mandatory and jurisdictionally broad. Reuters reported on July 1, citing the Financial Times, that the voluntary standards would set benchmarks for advanced models and clarify access rules both inside the United States and internationally. The divergence between US and EU approaches is creating a compliance fracture that enterprise customers building on American AI platforms must now factor into risk assessments.

The broader geopolitical stakes elevate the August framework announcement beyond a procedural milestone. According to reporting from Build Fast with AI, the Five Eyes intelligence alliance earlier warned that AI-powered cyberattacks are months away, sharpening the national security logic behind pre-release government review. The rapid improvement of lower-cost Chinese open-source models is simultaneously complicating Washington's ability to argue that restricting American frontier labs on security grounds serves the national interest if those restrictions push enterprise buyers toward foreign alternatives. Whether the framework due in early August can thread that needle — creating enough oversight to satisfy security agencies while maintaining enough openness to keep US labs competitive globally — will determine whether voluntary AI governance becomes a durable fixture of the technology industry or merely the first chapter of a longer regulatory escalation.