Washington Wants a Cut of AI, and Nobody's Ready

The Government as Shareholder

Here's the thing that keeps nagging at me about this week's news: the U.S. government wants to become a part-owner of AI companies. According to TechCrunch, President Trump said Friday he's been talking to AI executives about "concepts where pieces could be given to the American public, where the American public essentially becomes a partner with the companies." CNBC reported the administration has specifically been discussing an equity stake in OpenAI, with some of that equity potentially seeding a "Public Wealth Fund" the company recently proposed.

On its face, this sounds almost progressive. Distribute the gains of AI to ordinary citizens! Senator Bernie Sanders proposed something in the same neighborhood this week (a one-time 50% tax paid in stock by companies like OpenAI, Anthropic, and xAI). And Bloomberg reports that Sam Altman has been floating the idea of a government stake in major AI companies since early 2025. So you've got the far left and the Trump White House converging on the same basic impulse, which should make everyone a little nervous.

David Sacks, who recently stepped down as Trump's AI and crypto czar and now co-chairs the President's Council of Advisors on Science and Technology, posted that he understands why Sanders' version resonates "including with many on the right," but warned it would "accelerate the corporate-government fusion we're already sliding toward." He's not wrong. And the fact that this echoes the government's move to take a 10% stake in Intel last year suggests it's not a one-off impulse but a pattern. The government as venture capitalist is a role nobody voted for, and the conflict-of-interest questions are enormous. How do you regulate a company you own stock in? You don't. That's the point.

Safety Testing Without Safety Teams

OK so here's where it gets genuinely absurd. On Tuesday, Trump signed an executive order expanding voluntary safety testing of frontier AI models. Voluntary. As in, companies can participate if they feel like it. Ars Technica reports that critics are calling the order "short-sighted" and "performative," and honestly that might be generous.

The backstory is revealing. Trump had originally planned a big signing event with AI CEOs last month, then abruptly canceled it because (his words) the earlier version might have been a "blocker" to innovation. Reports indicated infighting between cybersecurity experts and officials committed to deregulation. The scrapped version would have given the government up to 90 days of early access to test models before deployment. The signed version? That window got narrowed significantly after industry pushback.

But the real problem isn't the watered-down timeline. It's that DOGE gutted the federal security teams who would actually do the testing. So you have an executive order that establishes a voluntary process, with no requirements for AI firms, to be carried out by teams that have been hollowed out by the administration's own cost-cutting initiative. It's like posting a "Lifeguard on Duty" sign at a pool where you fired the lifeguard last month. The order promises not "to stifle this innovation with overly burdensome regulation" and I believe it. Nothing about this will stifle anything.

Put these two stories together and a picture forms. The government wants equity upside from AI companies while simultaneously ensuring those companies face minimal oversight. That's not a policy. That's a term sheet.

The Revolving Door Spins Faster

Sriram Krishnan announced he's leaving his role as White House senior AI policy advisor at the end of June. He's the latest in a string of tech-industry-figures-turned-government-officials to cycle through the administration. Before the White House, Krishnan was a partner at Andreessen Horowitz and led product teams at Microsoft, Twitter, Yahoo, Facebook, and Snap. According to The Washington Post, he's planning to start an outside institution that will still give him a role in influencing Trump's AI policy.

Read that again. He's leaving government to build a private institution that shapes government AI policy. This is Washington's favorite magic trick: you serve for 18 months, build the relationships, then monetize the access from the outside while calling it "institution building." Krishnan highlighted the administration's AI Action Plan, which (and I'm paraphrasing only slightly) prioritized data center construction over regulation and safety. His departure follows Sacks' own exit from the czar role earlier this year.

What's striking is how Krishnan framed his accomplishments. Executive orders around AI, one that seeks to challenge state-level AI regulations, another on oversight that was "delayed and narrowed after industry pushback." He listed these as wins. The pattern is consistent: every time a guardrail appeared, it got sanded down. And now the people who did the sanding are leaving to build private organizations that will continue the work. It's a very efficient system, if you're an AI company.

Meanwhile, in Actual Science

Buried under all the policy maneuvering, something genuinely interesting happened. WIRED reported that Jeff Bezos is funding Flourish, a neuro AI company with $500 million in funding and a reported $2.5 billion valuation, co-founded by Thomas Reardon and Rob Williams (a former Amazon S-team executive). Their pitch is that current AI has dug itself into a hole: large language models are power-hungry, can't learn continuously, and have little in common with actual brains. A human brain runs on about 20 watts. A single chip in an AI training cluster uses more than 30 times that.

Flourish wants to build what they call "Cortex AI," a synthetic intelligence system designed to match the brain's computational capacity, learning efficiency, and power budget. It's wildly ambitious (and yes, I know how that sounds coming right after three sections of political cynicism). But the contrast is instructive. Washington is arguing about who gets equity in today's AI companies. Flourish is betting that today's AI architecture is fundamentally limited. Both can't be right. If the transformer-based paradigm really does hit a wall on power and continuous learning, then the government's equity stake in OpenAI might age about as well as a stake in Blockbuster circa 2005.

And then there's Microsoft, quietly rolling out "computer use" as a generally available feature in Copilot Studio as of May 2026, letting agents automate web and desktop apps by controlling browsers on behalf of users. It's the kind of incremental-but-significant move that gets lost when everyone's arguing about stock ownership and executive orders. Agents that can actually operate your computer. That's not a press release, that's a product.

The Upside Nobody's Earning

The through-line this week is a government that wants to participate in AI's financial upside while systematically dismantling its ability to manage the downside. Equity stakes without oversight. Safety orders without safety teams. Policy advisors who leave to build private influence shops. It all fits together a little too neatly.

What sticks with me is Sacks' phrase about "corporate-government fusion." He meant it as a warning about Sanders' proposal, but it describes the administration's own trajectory just as well. When the government owns stock in the companies it's supposed to regulate, and the people writing AI policy rotate between venture capital firms and the West Wing every 18 months, the fusion is already here. The only question is whether anyone will notice before the returns get distributed.