The Trump administration is preparing an executive order on AI and cybersecurity that would establish a voluntary framework for developers of advanced 'frontier' models to provide government agencies early access—at least 90 days before public release—according to multiple reports. Axios first detailed the draft, which focuses on bolstering defenses around critical infrastructure while creating a multi-agency review process to classify and assess 'covered frontier models' for risks, particularly in cybersecurity. Politico confirmed key elements, noting the order asks participating developers to engage with agencies including CISA, Treasury, and NIST, share model access, and provide select critical infrastructure providers preview capabilities. This arrives amid shockwaves from Anthropic's Claude Mythos Preview, a model demonstrating unprecedented autonomous vulnerability discovery and exploitation across major operating systems and browsers, as documented in Anthropic's own technical preview and evaluations by the UK's AI Security Institute showing major leaps in multi-step cyber attack simulations.

While framed as non-regulatory to preserve U.S. 'America First' AI dominance and avoid slowing innovation against global competitors, this initiative exposes subtler state-level power consolidation over frontier technologies. It builds on the December 2025 White House executive action preempting conflicting state AI laws, which identified 'onerous' state regulations and tied federal broadband funding to alignment with national policy—effectively centralizing governance in Washington to prevent a fragmented regulatory landscape. This federal preemption connects directly to broader AI governance battles: the tension between decentralized state-level experimentation (or restriction) and top-down national security imperatives.

The voluntary 90-day window functions less as optional oversight and more as an institutional bridge, fostering norms of intelligence-sharing between labs and the national security apparatus. In an era where models like Mythos can autonomously chain exploits at speeds outpacing human defenders, early visibility allows the state to prepare defenses, advise on mitigations, or potentially shape deployment—without the overt regulatory burdens of the prior administration's approach. Yet this entanglement risks normalizing government pre-review of transformative capabilities, blurring lines between private innovation and state-directed dual-use technology. Critics might see echoes of historical power grabs where 'voluntary' frameworks evolved into de facto standards, especially as cyber risks escalate from defensive tools to potential offensive autonomous agents. By tying this to Pentagon security, cyber hiring reforms, and threat-sharing, the order signals awareness of dual-use dangers while prioritizing U.S. competitive edge. The deeper connection others miss lies in how this preempts not just foreign adversaries but domestic policy diversity, consolidating authority over the trajectory of intelligence that could redefine cybersecurity, economic power, and even autonomous systems.