Washington’s trial runs are over. Months after mocking the previous administration’s AI safety rules, the Trump White House is scrambling to implement a nearly identical review process for frontier models. The pivot followed a classified briefing on Anthropic’s Mythos model, which specializes in finding and daisy-chaining software vulnerabilities. Officials who called pre-release testing 'communist' are now drafting executive orders to mandate it.
The urgency is data-driven. Palo Alto Networks CEO Nikesh Arora revealed that using models like Mythos and GPT-5.5 Cyber, his team found 26 critical exploits in a window where they typically find under five - a spike of over 700%. According to Arora, the time from breach to data exfiltration has collapsed from days to minutes, rendering the traditional 90-day patch window obsolete.
"The time-to-breach has collapsed from days to minutes. Arora argues that the traditional 90-day window for responsible disclosure is essentially dead."
- Casey Newton, Hard Fork
This creates a chaotic regulatory reality. Policy expert Dean Ball calls it an “informal, highly improvised licensing regime.” The state is asserting direct control, recently intervening to block the broad release of Anthropic’s Mythos, citing national security and compute constraints. The Pentagon simultaneously designates Anthropic a supply chain risk while using its models to scan for vulnerabilities.
The scale of AI infrastructure demands is triggering a separate policy reversal. A proposed data center in rural Utah is projected to draw nine gigawatts - more than twice the state’s current total usage. Tucker Carlson argues this forces elites to abandon 15 years of climate orthodoxy as energy infrastructure shifts to serve processors, not people.
"AI's massive power hunger forces elites to abandon 15 years of climate orthodoxy for fossil fuels."
- Tucker Carlson, The Tucker Carlson Show
Globally, the strategy is moving from rhetoric to industrial land grabs. Under Secretary of State Jacob Helberg outlined a deal securing a 4,000-acre industrial hub in the Philippines under American legal protection. This “forward-deployed industrial base” is a prototype for drawing private capital to build supply chains outside China’s reach, targeting monopolies in robotics and mineral refining.
The underlying tension is between unchecked acceleration and controlled deployment. Builders face compute scarcity and soaring costs, governments see a weapon, and the public sees a threat to jobs and purpose. The administration’s reactive turn to vetting suggests the technology’s lethality outpaced its libertarian politics.
