Trump Signs Executive Order on AI Innovation and Security

In a marked change from his largely hands-off approach to AI, President Trump signed an executive order on June 2, 2026 to address cybersecurity risks associated with frontier AI models. The executive order, titled Promoting Advanced Artificial Intelligence Innovation and Security (the Executive Order), establishes a voluntary partnership among the federal government, AI developers, and critical infrastructure providers to identify and remediate software vulnerabilities. The Executive Order encourages—but does not require—AI model developers to grant the federal government and "trusted partners" early access to models before release. The Executive Order expressly disclaims any intent to create a mandatory licensing, preclearance, or permitting regime for AI model development, publication, release, or distribution.

President Trump signed the order less than two weeks after he declined to sign a very similar proposed order due to concerns about stifling AI innovation, but the only substantive difference between them is the length of time provided to the federal government for early access to AI models. The earlier proposed order had a pre-release access period of 90 days; the Executive Order President Trump signed shortens that to 30 days.

The Executive Order arrives amid serious concerns about nation-state adversaries and other malicious actors using AI tools to launch cyberattacks at speeds and scale far beyond the capabilities of human hackers—and beyond what existing cyber defenses can effectively stop. The recent release of powerful cybersecurity-focused AI models, which have shown the ability to identify thousands of previously unknown software vulnerabilities, has compounded these concerns.

The Executive Order is a notable turn away from the broader trend of the Trump Administration's approach to AI issues, which began with a repeal of extensive, detailed AI-related executive orders from the Biden Administration. That approach continued with additional executive orders focused on encouraging AI development and innovation to ensure that the U.S. "wins the AI race" by preempting or penalizing burdensome state AI statutes and regulations. While the new Executive Order does not formally impose any obligations on AI developers, it does reflect a recognition that some degree of government involvement—if not formal oversight—in the release of powerful frontier AI models may be necessary. Notably, the Executive Order is silent as to key policy priorities the Administration outlined as recently as six months ago, including establishing a framework for preempting state AI laws, excluding so-called "woke" AI systems from use by federal agencies, or exploring the need for a national "reporting and disclosure standard" for AI models.

Although this Executive Order is framed as establishing "voluntary" programs, questions remain about whether, or how, the federal government many use its early access review timeline to impose soft mandates or leverage concessions from AI model developers. That review will be based on a classified benchmarking process and thresholds that will determine whether an AI model should be designated a "covered frontier model". Such designation may limit transparency and provide the federal government some discretion to expand or narrow the scope of its review without public notice or consultation.

Overview of the Executive Order

The Executive Order directs federal agencies to undertake four primary initiatives. Key directives include:

• Secure "Frontier Model" Deployment (Section 3). Within 60 days, the Department of the Treasury (Treasury), working with the National Security Agency (NSA), and the Cybersecurity & Infrastructure Security Agency (CISA), must develop a classified benchmarking process for assessing the advanced cyber capabilities of AI models and setting the threshold at which a model is to be designated a "covered frontier model." Treasury, NSA, and CISA then must design a voluntary framework through which AI developers may—but are not required to—provide the government with early access to a covered model for up to 30 days, "subject to appropriate confidentiality, cybersecurity, insider-risk, and intellectual-property protection, use, and nondisclosure requirements," before the model is released to other trusted partners. Developers also may collaborate with the federal government to select trusted partners who will have early access to covered models "to promote secure innovation and strengthen the cybersecurity of critical infrastructure." Assessment of covered AI models under the Executive Order may build on existing evaluations performed by the National Institute of Standards and Technology (NIST) through its Center for AI Standards and Information (CAISI). CAISI has entered into voluntary agreements with major AI developers to evaluate AI models prior to release for national security risks, with a focus on risks relating to cybersecurity, biosecurity, chemical weapons. This model testing process was originally established under the Biden Administration to mirror AI "Safety Institutes" in the U.K. and E.U., which took a more proactive approach to AI regulation and established direct oversight and review of cutting-edge frontier models. Given the significant concentration of technical expertise within CAISI (and other NIST organizations), we expect they will play a meaningful role in going forward reviews.
• Upgrading Federal and Critical-Infrastructure Systems (Section 2). CISA must issue Binding Operational Directives (BODs) to expedite defense of civilian federal systems, expand federal use of AI-enabled defensive tools, and facilitate access to AI-enabled cybersecurity tools—including the "covered frontier models" described in Section 3—to federal agencies, state and local governments, and critical infrastructure operators such as rural hospitals, community banks, and local utilities. The Committee on National Security Systems and the Department of War also are directed to prioritize the cyber defense of their systems within 30 days of the order.
• Vulnerability Detection (Section 2). Treasury, NSA, and CISA must stand up an "AI cybersecurity clearinghouse" to coordinate a voluntary program of vulnerability scanning, validation, and patch distribution with the AI industry and critical infrastructure operators. Although details on the clearinghouse are sparse, the Executive Order may envision participants leveraging AI tools to identify software vulnerabilities in critical infrastructure systems and coordinating efforts to remediate those vulnerabilities. The clearinghouse appears similar in concept to Anthropic's Project Glasswing, in which Anthropic has granted initial access to its Claude Mythos model only to a small group of critical infrastructure companies. Those companies reportedly have used Mythos to identify thousands of vulnerabilities in widely used software, allowing for remediation of those vulnerabilities before they are publicly reported. The order also directs the Office of Management and Budget (OMB) to assess whether existing federal grant funding can be directed towards developers of advanced AI vulnerability detection tools, and to expand the United States Tech Force Information Cybersecurity Specialist hiring and placement program.
• Criminal Enforcement (Section 4). The Executive Order directs the Attorney General to prioritize enforcement of existing federal criminal statutes, including the Computer Fraud and Abuse Act (18 U.S.C. § 1030) and prohibitions on identity-fraud (18 U.S.C. § 1028) and wire-fraud (18 U.S.C. § 1343), against anyone who uses AI to illegally access or damage a computer without authorization, including employing AI agents to unlawfully access data or information that is subsequently used for a criminal or unlawful purpose.

Broader Context: Rising Concern About Frontier AI Cyber Risk

The Executive Order lands amid intensifying concern that nation-state adversaries and other malicious actors can use AI to launch devastating cyberattacks to commit espionage, steal sensitive data, and disrupt critical infrastructure systems, among other serious harms. These concerns are no longer theoretical: Researchers have documented multiple cybersecurity attacks where threat actors used AI to automate large components of the attack with little human oversight. CrowdStrike's 2026 Global Threat Report documents an 89% year-over-year increase in attacks by AI-enabled adversaries. Earlier this month, Google's Threat Intelligence Group (GTIG) announced that it had for the first time identified a threat actor exploiting an unpublished vulnerability using an exploitation technique developed by AI.

Much of this concern has been driven by the arrival of highly capable AI models focused on coding and cybersecurity functions. While the human-language capabilities of advanced LLMs are often prominent in the public eye, one of the largest sources of recent AI revenue growth has been the rapid adoption of AI coding tools by companies seeking to supplement (or replace) human coders with AI. Given the ever-improving capabilities of AI in coding, it is not surprising that AI models have demonstrated the ability to find multitudes of previously unreported vulnerabilities—some of which are decades old—in web browsers, operating systems, open-source libraries and other software used around the world. AI models have also successfully developed working exploits (software and techniques to exploit security vulnerabilities) for some vulnerabilities far faster than humans could, compressing the window between discovery and exploitation toward real time. Both Anthropic and OpenAI recently have released cybersecurity-focused models in restricted fashion to vetted partners.

Looking Ahead

AI developers and critical infrastructure operators should closely monitor implementation of the Executive Order and its voluntary partnerships. Many specific questions, such as who will be able to participate in the "AI cybersecurity clearinghouse," what legal and confidentiality protections the government will offer during the period of "early access" to AI models, how the government will define a "covered frontier model," and how "trusted partners" for early access will be selected, are still to be answered. DWT's Privacy & Security and Artificial Intelligence teams will continue to monitor these developments.

+++

Michael Borgia and K.C. Halm are partners in DWT's Washington, D.C. office, Andrew Lewis is counsel in the firm's San Francisco office, and Christopher Savage is a partner in our Washington, D.C. office. For questions or more insights, please reach out to the authors or another member of our privacy & security and artificial intelligence teams and sign up for our alerts.