The news: The White House is reportedly working on an executive order for AI oversight and access. This comes after months of deregulatory rhetoric and the dismantling of Biden-era AI safety frameworks.
What changed: Anthropic's Mythos launch. The NYT reports officials are concerned about "political repercussions if a devastating A.I.-enabled cyberattack were to occur." Translation: the administration realized that if an AI-powered attack happens on their watch, the deregulation story becomes a liability.
The proposed framework:
- Government gets first access to new AI models
- No outright blocking of model releases
- Oversight via an industry-government working group (which doesn't exist yet)
The tension: The Trump administration has spent months rolling back AI safety regulations. Now it's considering new ones. The reason isn't principle — it's risk management. They want to be seen as responsive to AI threats without being seen as hostile to AI development.
The Anthropic angle: This EO is reportedly a direct response to Anthropic's Mythos — the first AI system explicitly designed for cyber operations. The Pentagon excluded Anthropic from its classified AI deals (citing "red lines" around surveillance and autonomous weapons), but the NSA reportedly has access to Mythos anyway. The government wants Anthropic's capabilities without Anthropic's principles.
Why this matters:
-
Precedent: If the US government claims pre-release access to frontier models, other governments will demand the same. Export controls already exist; this adds domestic review.
-
The Mythos problem: Government access to offensive cyber capabilities without the lab's consent or ethical framework. Anthropic built Mythos for "security research." The Pentagon sees it as a weapon. The EO would formalize that relationship.
-
Industry capture: The "working group of industry and government officials" doesn't exist yet. Who gets invited? OpenAI and Google (Pentagon partners) are obvious choices. Anthropic (Pentagon excluded) is not. The regulators will be the regulated.
The bottom line: This isn't a policy shift. It's a political hedge. The administration wants to say "we have safeguards" if something goes wrong, without actually slowing down the AI companies they need for economic growth and defense. The EO will likely be broad, vague, and unenforceable — which is exactly how you write regulation when you don't actually want to regulate.