Trump Administration Prepares Voluntary AI Model Review Framework

Key Takeaways

• White House plans Thursday AI executive order for voluntary government reviews of new AI models.
• Anthropic and OpenAI are engaging with officials on a potential 90-day or 14-day review period.
• Framework aims to mitigate cybersecurity risks by identifying vulnerabilities in unreleased advanced AI models.
• Recent shifts include Anthropic’s Mythos model exposing unprecedented cybersecurity exploitation capabilities.

The White House is preparing to issue a significant AI executive order as early as Thursday, aimed at establishing a voluntary government review process for new artificial intelligence models before they are released to the public. According to sources with knowledge of the situation, this directive will compel AI companies to share their advanced models with the government for a specific period ahead of their launch, marking a distinct shift in federal engagement with the tech sector.

The timeline for these reviews has been a primary point of negotiation between industry leaders and government officials. One draft of the order proposes a 90-day pre-launch review period, a timeframe that allows for thorough analysis but may be perceived as lengthy by some companies. Conversely, industry representatives, including those from Anthropic and OpenAI, have expressed a preference for a shorter window, such as 14 days, to balance security needs with rapid deployment cycles. These discussions indicate a collaborative, albeit complex, effort to harmonize regulatory oversight with the fast-paced nature of AI development.

Advanced AI models possess the potential to supercharge cyberattacks, a risk that early government review aims to mitigate. Experts warn that unreleased models could harbor vulnerabilities that, if exploited, could threaten national security and public safety. By implementing a voluntary framework, the government seeks to identify and neutralize these threats before they are unleashed in the world. One draft version of the order is divided into two main sections: one addressing cybersecurity and another defining "covered frontier models" eligible for this early review process.

The cybersecurity portion of the order outlines the creation of a voluntary "clearinghouse" facilitated by the Treasury Department, other agencies, and participating AI companies. This clearinghouse is designed to find and fix security vulnerabilities in unreleased AI models. Additionally, the order calls for increased hiring at the US Tech Force, a specialized body of engineers tasked with modernizing government computer systems, further strengthening the federal capacity to evaluate and secure critical infrastructure.

The Trump administration has historically taken a hands-off approach to AI regulation. However, recent developments have prompted a change in strategy. Anthropic recently unveiled its Mythos model, which it claims can exploit cybersecurity vulnerabilities at an unprecedented pace. This revelation has heightened concerns among officials regarding the potential misuse of advanced AI capabilities. In response, Anthropic has chosen not to release the model publicly but is instead granting access to a tightly controlled consortium of companies through its Project Glasswing initiative. The company is working closely with federal, state, and local representatives to ensure responsible usage.

Similarly, OpenAI has begun granting businesses and governments special early access to its latest AI models to help shore up cyber defenses. This proactive measure aligns with the broader goals of the proposed executive order, which seeks to foster cooperation between the government and the private sector. The Department of Commerce’s National Institute of Standards and Technology (NIST) announced earlier this month that major tech companies would share unreleased versions of their AI models with the government for national security and public safety-related evaluation. However, the announcement regarding this sharing initiative is no longer available on the Commerce Department’s website, suggesting potential internal revisions or strategic shifts in communication.

White House Signals Shift in AI Oversight Strategy

The White House has declined to comment on the potential release of the executive order, maintaining its standard posture of non-commenting on unannounced administrative actions. Nevertheless, the involvement of major industry players like OpenAI and Anthropic in these discussions underscores the urgency and importance of the issue. The proposed framework represents a critical juncture in the ongoing dialogue between the federal government and the AI industry. If successful, this voluntary review process could set a precedent for future regulatory approaches, potentially reducing the risk of malicious actors exploiting unreleased AI technologies. As the administration prepares to move forward, the focus will remain on balancing innovation with security, ensuring that advancements in AI are harnessed for public good while mitigating potential threats. The outcome of these negotiations will likely influence not only domestic policy but also global standards for AI safety and cybersecurity.