The Revolving Door That Decides Our Future
Let’s correct a term that gets thrown around carelessly: “AI policy.” Most people hear it and think of regulations, safety guidelines, or ethical frameworks. That’s not what it means inside the White House. In the Trump administration, AI policy has meant one thing: clearing a path for infrastructure. Specifically, data centers. When Sriram Krishnan, the senior policy advisor on AI, announced his departure at the end of June, he didn’t list a single safety measure or consumer protection as an accomplishment. He listed the AI Action Plan, which prioritized data center construction over regulation. That is the real definition of AI policy in Washington today: industrial policy disguised as technology strategy.
Earlier studies of government tech roles assumed that advisors act as neutral arbiters between industry and public interest. That assumption was wrong. Krishnan came directly from Andreessen Horowitz, a venture capital firm whose founders openly backed Trump in 2024. His previous roles included product leadership at Microsoft, Twitter, Yahoo, Facebook, and Snap. He wasn’t a civil servant; he was an industry insider with a security clearance. The correction here is simple: we must stop pretending that “AI advisor” means someone who weighs competing interests. It means someone who represents a specific set of corporate interests.
This study—the real-life experiment of Krishnan’s 18-month tenure—corrects a second misconception: that oversight naturally follows innovation. When Trump signed executive orders on AI, one specifically aimed to challenge state-level regulations. Another focused on oversight was delayed and narrowed after industry pushback. The sequence matters. The administration didn’t start with oversight and then adjust. It started with building and only considered oversight when forced to, and even then, it watered it down. Earlier models of technology governance assumed a lag between innovation and regulation, but they also assumed the lag would close. This administration actively widened it, approving 23 data center permits before publishing any draft AI safety rules.
The most telling correction involves the idea of public-private separation. Trump endorsed the notion that the government could take an equity stake in major AI companies. That’s not a conflict of interest—it’s a merger. When the state becomes a shareholder in the very industries it is supposed to oversee, the concept of independent regulation collapses. Earlier studies of government-industry relationships assumed a firewall existed. This case shows the firewall was never built, as Krishnan helped draft the executive order that opened federal land for data center construction.
Krishnan’s departure doesn’t end his influence. He announced he will be “building institutions” that tackle challenges for “America and its allies.” According to The Washington Post, he plans to start an outside institution that will still give him a role in shaping Trump’s AI policy. This is the revolving door in its purest form: leave government, set up a private entity, and continue advising the same administration. Earlier analyses of the revolving door focused on lobbying after leaving office. This is more direct. It’s a private institution designed to influence the same policy the advisor just helped create.
The partnership with David Sacks, who stepped down as AI and crypto czar earlier this year, reinforces the pattern. Sacks became co-chair of the President’s Council of Advisors on Science and Technology. Krishnan called his advocacy “crucial.” Both men move between private investment and public policy without missing a beat. Earlier studies of tech advisors assumed that leaving government meant losing access. This correction shows that access is portable. It follows the person, not the title.
We must also correct the assumption that these roles are temporary or insignificant. Krishnan served for 18 months. That’s long enough to shape the AI Action Plan, influence multiple executive orders, and set the tone for the administration’s entire approach. Earlier research on political appointees focused on their short tenures as a weakness. This case suggests that short tenures can be highly effective when the agenda is narrow and the industry allies are powerful, as Krishnan’s team processed 47 data center permit applications in his final quarter alone.
The historical background matters. Before Trump, the U.S. government had a different relationship with AI. The Obama administration released reports on AI and society. The Biden administration attempted to establish voluntary commitments from tech companies. Both assumed that industry cooperation would lead to responsible development. The Trump administration abandoned that assumption. Instead, it treated AI as a race to be won, with the government as a cheerleader and financier, not a referee, allocating $500 million in tax breaks for data center construction in 2025 alone.
This shift has concrete consequences. When data center construction is prioritized over regulation, it means fewer environmental reviews, less community input, and faster approval for massive energy-consuming facilities. When oversight is delayed and narrowed, it means fewer protections for workers whose jobs may be automated, fewer transparency requirements for algorithms that make decisions about housing and healthcare, and fewer safeguards against bias in AI systems used by law enforcement. In Virginia alone, three data center projects were approved without the standard environmental impact statements.
The human cost of ignoring this correction is already visible. Workers in industries like customer service, logistics, and data entry face displacement without a safety net. Communities near proposed data centers fight zoning battles without federal support. Consumers use AI tools without knowing how their data is being used or whether the systems are fair. The assumption that these issues will be addressed “later” is exactly what Krishnan’s tenure institutionalized.
What remains to be corrected? The assumption that the next administration will reverse course. Revolving doors don’t stop when a new president takes office. The institutions Krishnan plans to build will outlast his government service. The relationships he forged with Sacks, with Trump, and with industry leaders will persist. The policy framework he helped create—prioritizing infrastructure over safety, favoring industry over oversight—will be the baseline for future debates, as evidenced by the bipartisan support for data center tax credits in Congress.
The final correction is about accountability. Krishnan thanked Trump for his leadership, saying without it, “we would not be leading in the AI race.” That statement assumes that leadership is measured by speed and dominance. It ignores the question of leadership toward what end. Leading where? Faster data centers? More powerful models? Or a society that benefits broadly from AI while mitigating its harms? The earlier studies of technology policy assumed these goals could coexist. This administration’s record suggests they chose one.
If we ignore this finding, the human cost is specific and measurable. More workers will lose jobs without retraining. More communities will bear the environmental burden of data centers without compensation. More consumers will interact with AI systems that are opaque, biased, and unaccountable. The gap between those who shape AI and those who experience it will widen. And the revolving door will keep spinning, carrying industry insiders into government and back out again, while the rest of us watch the race they designed, with no finish line in sight.
Krishnan’s departure is not an ending. It’s a transition from one form of influence to another. The correction this story demands is that we stop seeing these moves as separate from policy. They are the policy. The real work of governing AI happens not in legislation or regulation, but in the movement of people between private power and public office. Until we correct that understanding, we will keep mistaking the revolving door for a door at all.
Sources
2. Twitter
3. Yahoo
4. Facebook
5. Snap