Sriram Krishnan Exits White House AI Role to Launch New Policy Institution
Sriram Krishnan is stepping down as the White House's senior AI policy advisor. He's reportedly launching a new institution aimed at continuing to shape the Trump administration's approach to AI governance.
Original sourceSriram Krishnan, who served as a senior policy advisor on artificial intelligence at the White House Office of Science and Technology Policy, is departing his role. Krishnan, a former partner at a16z and a longtime figure in the tech-policy corridor, was one of the more recognizable Silicon Valley names to take a position in the Trump administration's AI apparatus.
Rather than returning to the private sector, Krishnan is reported to be founding a new institution focused on AI policy — one that will ostensibly continue influencing the direction of federal AI strategy from outside the executive branch. Details on funding, structure, and scope of the organization have not been publicly confirmed.
His departure comes at a pivotal moment: the U.S. is still navigating the tension between aggressive AI deregulation and the need for baseline safety frameworks, and the administration's posture toward China on AI competition remains a live policy front. Whether a quasi-independent institution can hold real influence over White House priorities — as opposed to operating as a think tank with proximity but no authority — remains the central question.
Krishnan's move continues a pattern of tech-adjacent figures cycling between government advisory roles and external organizations, attempting to maintain policy leverage without the constraints of federal employment. The structure and independence of his new institution will determine whether this is a substantive continuation of his advisory work or a dignified exit with a mission statement attached.
Panel Takes
The Skeptic
Reality Check
“'New institution to shape AI policy' is a phrase that has launched a thousand think tanks with zero measurable impact. The question isn't whether Krishnan has influence — he clearly does — it's whether an externally-funded organization can move faster than lobbyists and slower than actual legislation to matter. I'd want to see the funding sources and the first concrete policy output before calling this anything other than a soft landing with good branding.”
The Futurist
Big Picture
“The bet here is that U.S. AI policy will increasingly be written by a small, networked class of technologists who rotate between government, VC, and quasi-public institutions — and that formal titles matter less than access and credibility with the people who hold them. That's a plausible thesis, and if Krishnan's institution becomes the place where the administration stress-tests its AI frameworks before announcing them, it's genuinely infrastructure. The dependency is that the current administration stays engaged on AI substance rather than just AI optics — which is not guaranteed.”
The Founder
Business & Market
“Who funds this matters more than anything else Krishnan will say about its mission. If it's funded by the same AI companies whose products will be shaped by the policies it influences, the institution's credibility has a ceiling that no amount of talent can raise. The model only works if the funding structure is defensible — endowment, government grants, or sufficiently diversified private donors — because the moment a hyperscaler's logo is on the annual report, the policy output gets discounted to zero by anyone paying attention.”
The PM
Product Strategy
“The job-to-be-done here is 'translate technical AI reality into federal policy at speed' — and the honest question is whether an outside institution can do that better than a seat at the table. Krishnan had the seat; he's now betting the institution is a better product. That only works if the administration has a consistent habit of consulting it, which is a distribution problem, not a mission problem. Until there's a clear mechanism for how this institution's work actually gets into decision-making pipelines, the roadmap and the shipped features are two different things.”