Trump signs executive order creating voluntary AI model sharing with 30-day government access
Trump signs executive order creating a voluntary AI model-sharing framework that gives the U.S. government 30 days’ access to advanced AI models and establishes an AI cybersecurity coordination centre.
President Donald Trump signed an executive order on Tuesday establishing a voluntary AI model sharing framework that allows developers to grant government access to advanced systems before public release. The move introduces a 30-day window for pre-release review and coordination, making “AI model sharing” a central element of the new White House approach to artificial intelligence safety and security. Administration officials said the order seeks to balance national security and critical infrastructure protection with preserving the pace of private-sector innovation.
White House unveils voluntary AI model-sharing framework
The executive order invites leading AI companies to voluntarily share their most capable models with designated U.S. agencies ahead of launch. Participating developers, including major research labs and commercial providers, would provide model access to specified government teams to enable vulnerability assessments and risk mitigation planning. The White House framed the initiative as cooperative and non-mandatory, emphasizing voluntary partnerships rather than direct regulatory compulsion.
Companies that participate will not be compelled to release proprietary code or data beyond what is necessary for review, according to the outline circulated within the administration. Officials described safeguards intended to protect intellectual property while enabling the government to evaluate potential threats to sectors such as finance, healthcare and critical infrastructure. The framework is explicitly aimed at early detection of misuse, exploitation and systemic vulnerabilities.
Thirty-day government access agreed as compromise
The order sets a 30-day period during which the government may access advanced models before their public rollout. That timeframe was presented as a compromise after an earlier leaked draft reportedly proposed up to 90 days of advance review, while technology firms lobbied for as little as 14 days. The 30-day window is intended to give agencies sufficient time to surface significant cybersecurity or safety issues without imposing overly long commercial delays.
Administration spokespeople said the timeline balances the need for prompt technical appraisal with the commercial realities of rapid product cycles in AI development. The voluntary nature of the program means companies retain discretion over whether to participate, but officials signaled that collaboration could become a competitive expectation for market leaders. Analysts say the practical effect will depend on how agencies use the access period and whether technical reviews produce actionable findings.
Silicon Valley reaction and innovation concerns
The announcement prompted mixed reactions from technology leaders and investors, with some welcoming greater coordination and others warning against burdens that could slow innovation. David Sacks, a Silicon Valley investor who serves in a Trump administration role focused on AI and digital assets, publicly cautioned that excessive regulation could hinder American competitiveness. He argued on social media that removing bureaucratic obstacles is essential for winning the global AI race.
Industry groups stressed the need for clear legal protections and strict limits on governmental use of proprietary models, noting that disclosure requirements could affect intellectual property and commercial advantage. Some companies indicated they would evaluate participation on a case-by-case basis, factoring in the scope of review, legal safeguards and potential impacts on go-to-market timelines. Observers say the administration will need to work closely with the private sector to maintain trust and ensure meaningful cooperation.
Creation of an AI cybersecurity coordination centre
The executive order directs the Treasury Department, the National Security Agency and the Cybersecurity and Infrastructure Security Agency to establish a coordinated AI cybersecurity centre. That body will operate in voluntary partnership with developers and operators of critical infrastructure to identify software vulnerabilities and accelerate remediation. Officials described the centre as a hub for sharing technical findings, coordinating patching efforts and disseminating threat intelligence specific to AI systems.
The centre is tasked with streamlining vulnerability disclosure between private firms and government teams, with an emphasis on speed and technical rigor. By centralizing coordination, the administration intends to reduce duplicate efforts across agencies and provide operators with prioritized guidance when serious risks are identified. The order also calls for ongoing collaboration with infrastructure owners to protect systems such as banking, healthcare and utilities from AI-driven exploitation.
Anthropic model ‘Mythos’ and safety red flags prompted action
The policy shift comes after concerns about an advanced model developed by Anthropic, referred to by some reports as “Mythos,” which the company chose not to release publicly because of its capacity to identify system vulnerabilities. That decision helped crystallize worries in government circles about the potential for powerful models to discover and weaponize flaws in critical systems. Officials said the order is partly a response to that episode and designed to provide a structured path for evaluating similar future developments.
Administration sources indicated that the Anthropic case demonstrated both the technical reach of next-generation models and the urgency of establishing mechanisms for early assessment. The voluntary sharing framework is intended to give government teams visibility into models whose capabilities could pose substantial risks, while avoiding heavy-handed mandates that might drive innovation offshore. Legal and technical experts say clear rules of engagement will be crucial to prevent adversarial misuse while respecting commercial confidentiality.
Policy reversal and next steps for U.S. AI governance
The executive order also revoked a prior Biden-era directive on AI oversight, a move the White House said was intended to create a single, updated approach aligned with the current administration’s priorities. Officials emphasized a preference for partnership and coordination over prescriptive regulation, though they acknowledged that future steps could include formal rules if voluntary cooperation proves insufficient. The administration has set a timeline for agencies to implement the AI cybersecurity centre and issue further guidance to participating firms.
Legal analysts expect follow-up guidance on the scope of government access, data handling protections and liability limits for model developers who volunteer information. International observers will also watch how the U.S. balances security, innovation and market competitiveness, particularly in relation to China’s AI ambitions. For now, the voluntary AI model sharing framework marks a significant shift toward structured public-private cooperation in managing the risks posed by powerful AI systems.
The order signals a new chapter in U.S. AI policy that prioritizes early technical review and improved incident coordination while attempting to preserve the speed of commercial development. How companies respond and how quickly the coordination centre delivers actionable protections will determine whether the initiative reduces systemic risk without compromising American leadership in the technology.
