Trump Sprints on AI Oversight: New 30-Day Federal Mandate
President Donald Trump has signed an updated executive order establishing a 30-day window for federal safety reviews of frontier AI models. This represents a radical reduction from the previously discussed 90-day period. Industry heavyweights like OpenAI’s GPT-5.5 and Anthropic’s Claude Mythos are now directly in the crosshairs: developers must grant the government exclusive access to model weights and architectures prior to any official release. The administration is attempting a delicate balancing act—identifying critical cybersecurity vulnerabilities without saddling American corporations with a "regulatory anchor" that could stall them in the global race.
The shortened deadline emerged from intense political horse-trading. The initial draft was shelved following pushback from industry leaders and former "AI Czar" David Sacks, who argued that a three-month delay would be a death sentence for innovation. Consequently, White House Chief of Staff Susie Wiles, Treasury Secretary Scott Bessent, and Cybersecurity Director Sean Cairncross revived the document by slashing the review period by two-thirds. Anthropic was quick to endorse the move, calling it an "important step for U.S. leadership"—corporate-speak for relief that regulators have accepted a pace dictated by the market rather than safety protocols.
Beyond domestic oversight, this decision gives Scott Bessent more leverage to resume cross-border regulatory negotiations with China. The White House is betting that a 30-day technical sprint is enough to catch catastrophic bugs without turning the audit into a bureaucratic quagmire.
In our view, expecting deep alignment analysis and quality red-teaming in just four weeks is a dangerous illusion. Comprehensive safety testing requires time that developers simply don't have in an arms-race environment. Thirty days looks less like a genuine threat filter and more like a "speed tax" that Big Tech is willing to pay for the badge of safety, while state experts scramble to parse game-changing code on the fly.
New regulations target frontier models like GPT-5.5 and Claude Mythos. Developers must disclose model weights and architecture before public release. The federal audit window has been slashed from 90 to 30 days. The initiative aims to detect cyber threats without hindering development speed.