You’ve probably noticed the headlines. The White House is expanding its AI policy. New frameworks, new guardrails, new oversight. It all sounds very responsible, very measured, very much like adults finally stepping into the room.
But here’s the thing nobody’s saying out loud: the U.S. government is now deeply, structurally dependent on the exact same companies it’s pretending to regulate.
That’s not a policy. That’s a hostage situation with better branding.
Let me walk you through the tension that everyone in Washington feels but no one wants to name. The White House wants American AI labs to maintain global dominance. It also wants to put guardrails on frontier models that could reshape warfare, truth, and labor markets. Those two goals are already pulling in opposite directions. But the deeper conflict is this: the Pentagon, the intelligence community, and federal agencies are now actively consuming the products of OpenAI, Anthropic, and Google DeepMind at an accelerating pace. They need these models for national security. They need them for cyber defense. They need them to keep up with China.
You cannot regulate someone you desperately need to call at 2 a.m. when the missiles are in the air.
This is the conflict of interest that undermines every framework the White House publishes. When the same labs building your military’s AI backbone are the ones you’re supposed to be restricting, enforcement becomes a negotiation, not a rule.
Think about how this plays out in practice. A lab develops a frontier model that demonstrates dangerous capabilities. Maybe it can generate biological attack plans. Maybe it can automate cyber offensives at scale. The oversight framework says: report this, pause deployment, submit to review. But the lab also has a $200 million contract with the Department of Defense. The model is already integrated into classified systems. What happens then?
Does the government actually shut it down? Or does it quietly grant an exception because national security can’t wait for a safety audit?
We both know the answer.
Every AI safety framework the government writes contains an invisible asterisk that reads: unless we really need it.
The voluntary commitments we saw last year were already a joke. Companies promising to self-regulate is like a casino promising to police its own slot machines. Now the White House is moving toward binding oversight, which sounds tougher. But binding oversight requires enforcement, and enforcement requires leverage. When your regulator has less leverage than your customer, the regulation is theater.
Here’s where it gets genuinely uncomfortable. The national security argument that drives AI investment in this country is the same argument that makes meaningful regulation impossible. You can’t tell Congress, “We need to pour billions into AI to stay ahead of China,” and then turn around and say, “But we need to slow down the labs building it because the risks are too high.” Those messages cancel each other out in every appropriations hearing, every classified briefing, every policy room where decisions actually get made.
If you work in tech, this matters to you directly. The companies building the tools you use every day are being shaped by a policy environment that is fundamentally conflicted. The rules that emerge from this process won’t be designed to protect you. They’ll be designed to protect the government’s access to the models it needs while creating the appearance of public safety.
The real risk isn’t that AI will be unregulated. It’s that AI will be regulated by people who are too invested in it to regulate honestly.
What would actual oversight look like? It would require a regulatory body with no procurement relationship with the labs it oversees. It would mean firewalling national security AI procurement from civilian AI regulation entirely. It would mean accepting that some models might deploy slower in the U.S. than in countries with fewer scruples, and treating that as a feature, not a bug.
None of that is happening. What’s happening is a slow-motion merger between the federal government and a handful of AI labs, dressed up in the language of oversight.
The White House isn’t expanding AI policy. It’s expanding AI dependency and calling it governance.
And when the next frontier model does something none of us expected, the people who were supposed to be watching will be the same people who couldn’t afford to look away.
FAQ
Q: Isn't the government already regulating industries it depends on, like defense contractors?
A: Yes, but defense contractors don't ship consumer products used by hundreds of millions of people, and they don't iterate their core technology every few months. The speed and reach of AI labs makes this a fundamentally different dynamic.
Q: What does this mean for people working in tech or investing in AI?
A: Expect policy that looks tough but bends wherever national security procurement is involved. Companies with deep government contracts will face softer enforcement. Smaller players will bear the real regulatory burden. Plan accordingly.
Q: So you're saying AI regulation is pointless?
A: Not pointless — compromised. Regulation written by a dependent party isn't regulation. It's procurement policy wearing a safety costume. The framework matters, but only if the people enforcing it don't need the thing they're regulating.