WASHINGTON, June 3, 2026, 05:10 (EDT) – In a significant policy shift, the White House has issued an executive order requiring leading artificial intelligence developers to submit their most advanced models for a voluntary federal cybersecurity review lasting up to 30 days before broader deployment. The move directly impacts major players including OpenAI, Google, and Anthropic, whose cutting-edge systems now face potential launch delays and increased disclosure requirements.
The order, signed by President Trump, directs U.S. agencies to evaluate frontier AI models for security vulnerabilities, particularly those capable of identifying or exploiting software bugs in critical infrastructure such as banks, hospitals, and power grids. This review process, while voluntary, creates a new channel for government oversight that could alter product rollout calendars for companies that choose to participate as a business necessity.
Anthropic’s Claude Mythos Preview, a cyber-focused AI model designed to hunt for security flaws, has been at the center of the debate. The company is expanding access to this model through Project Glasswing, growing from approximately 50 organizations to about 200 across more than 15 countries. Sectors covered include healthcare, power, water, communications, and hardware. Anthropic reported in May that the project and its 50 initial partners identified over 10,000 high- or critical-severity vulnerabilities, many posing serious risk if exploited.
The executive order stops short of imposing mandatory licensing or pre-clearance requirements. The White House emphasized that the process is voluntary and does not establish a licensing regime for AI. However, it gives Washington a new channel to participate in pre-release testing, potentially affecting the timing of product launches for firms that opt to engage with the process as a business imperative.
Industry leaders have largely backed the initiative. Google’s Kent Walker called the move “an important step forward,” while OpenAI CEO Sam Altman said it “gets the balance right.” Cisco Chair and CEO Chuck Robbins noted that AI is “rewriting the rules of cybersecurity” and welcomed the order as a way for defenders to accelerate while preserving room for private innovation. Goldman Sachs CEO David Solomon, whose firm is testing Mythos, emphasized the importance of information-sharing so that other sectors can adapt.
South Korea’s Science Ministry announced Wednesday that the Korea Internet & Security Agency has gained access to Mythos through Project Glasswing, working with Anthropic. According to Reuters, Samsung Electronics, SK Hynix, and SK Telecom are included in the expanded initiative, underscoring the global reach of these AI cybersecurity efforts.
The White House is framing the order as a security measure rather than sweeping AI regulation. Its fact sheet calls for the establishment of an AI cybersecurity clearinghouse to share software bugs and fixes, along with providing AI-based cyber tools to federal agencies, state and local governments, and critical infrastructure operators. This clearinghouse would act as a central hub for disseminating vulnerability information.
However, voluntary rules are raising questions. Juan Londoño, a policy analyst at the Cato Institute, told AP that while the order is “a step in the right direction,” unclear guidelines on which models qualify and who receives trusted access could set a “dangerous precedent” if the government uses them against companies with which it is in conflict. For AI firms, the primary risk is not an outright ban but potential delays, extra disclosure, and federal reviews that could surface issues just as competitors are preparing to launch. For regulators, the challenge is that if the process remains optional, major models may bypass it, leaving Washington with a framework that top companies sometimes ignore.
