Trump pauses AI security order over China competition fears

The signing pens were ready, and the White House was just hours away from a public rollout. Then President Trump abruptly slammed on the brakes.

Trump confirmed to reporters in the Oval Office that he postponed a planned executive order aimed at frontier AI security. His explanation was blunt: he didn’t like certain provisions in the draft, and he worried that tying up domestic developers with government red tape would blunt America's competitive edge against rivals like China.

The drafted order, circulating through Washington over the past 24 hours, aimed to build a formal pre-release vetting pipeline for cutting-edge AI. Frontier labs would be required to share their models and relevant data with the federal government anywhere from 14 to 90 days before public launch. It would also open testing pipelines for cybersecurity defenders in critical infrastructure sectors like finance and healthcare.

Inside the 90-day sandbox

The draft dished out serious authority across federal agencies.

The National Security Agency was tapped to run classified vulnerability assessments. The Treasury Department was tasked with brokering information-sharing pacts between AI labs and critical infrastructure defenders. Meanwhile, agencies like CISA, ONCD, and NIST would define the compute thresholds for which frontier models qualified for scrutiny.

One former official familiar with the text noted that federal agencies are done watching vendor-led slide decks. In the past, companies held the government's hand and showed officials only what they wanted them to see. Today, agencies feel they finally understand the technology well enough to demand the raw models and test them themselves.

That urgency didn't come out of nowhere. Recent high-profile systems like Anthropic's Mythos and OpenAI's GPT-5.5 have triggered alarms after being flagged for potential cybersecurity risks.

Trapped by the "no guardrails" promise

The stakes on the ground have moved well past theoretical debates. Ram Shankar Siva Kumar, who founded Microsoft's AI red team, pointed out that his outfit started in 2019 with just a handful of security engineers. Today, it employs psychologists, linguists, and bioweapons specialists. From targeted military ops and autonomous drone swarms to state-sponsored hacking campaigns, frontier AI is already woven into global warfare.

Yet Trump's administration walked straight into a trap of its own making. After entering office loudly rejecting AI guardrails and promising total deregulation to outpace foreign rivals, officials suddenly realized that preventing frontier models from breaking national security requires the very guardrails they campaigned against.

By pulling the plug at the eleventh hour, Trump demonstrated just how hard it is to reconcile an aggressive deregulation stance with the terrifying realities of autonomous cyber threats.

Why it matters

In the AI world, 90 days isn't just a regulatory pause; it's practically an entire product generation. Forcing frontier labs to sit in a federal waiting room for up to three months while overseas rivals and open-source projects push updates weekly is a legitimate commercial nightmare.

Still, we think Trump's decision to freeze the order kicks a critical can down the road. Models like Mythos and GPT-5.5 have proven that frontier intelligence doubles as an elite cyber-weapon. Framing basic pre-deployment security audits solely as a handicap against China makes for great political rhetoric, but it leaves critical infrastructure exposed to automated exploits.

The White House clearly wants American AI to move at breakneck speed without accepting the catastrophic risks that come with moving blind. For builders and enterprise buyers, the compliance hammer hasn't dropped just yet, but the underlying tension isn't going anywhere.