Sriram Krishnan's planned exit does not look like a retreat from AI policy. It looks like a move into a less visible arena where Washington, Silicon Valley and the next generation of model rules will keep colliding.
Sriram Krishnan is leaving the White House at the end of June, but the larger story is where his influence may go next. The senior policy adviser for artificial intelligence has told officials he plans to create an outside institution focused on technology policy, a move that could keep him close to the Trump administration's AI agenda while taking him beyond the normal constraints of a government post.
According to The Washington Post, Krishnan helped shape the administration's pro-industry AI strategy, including the AI Action Plan, data center buildout priorities and federal procurement rules aimed at removing politically biased AI systems from government use. That is not a small portfolio. It is the operating map for how the federal government wants AI to grow, where it wants infrastructure built and how much room it wants companies to have before oversight begins to bite.
For founders, investors and AI labs, the timing matters. On June 2, President Trump signed an executive order creating a voluntary framework that would give the government access to certain powerful frontier models for up to 30 days before those systems are shared with trusted third parties. The order also directs agencies including Treasury, the National Security Agency, CISA and NIST to develop classified benchmarks for identifying models that may pose advanced cyber risks. That is a clear shift from the earlier deregulatory posture, even if the White House is careful to say it is not creating a licensing regime.
Outside institutions can be useful. They can gather technical talent, publish policy papers, convene companies and give government officials access to people who understand fast-moving markets better than a typical agency process does. In AI, that can be especially valuable because the technology changes faster than procurement schedules, congressional hearings or agency rulemaking.
But there is another side to it. When policy influence moves outside formal government, it becomes harder to see who is shaping the argument, who is funding the work and which companies stand to benefit. Krishnan came to the administration from Andreessen Horowitz and previously worked at Facebook and Twitter. Those connections are not disqualifying. They are also not irrelevant. AI policy is now deeply tied to capital allocation, cloud contracts, chip access, energy demand and the global fight over which technology stack becomes dominant.
Washington is no longer only saying move fast
The Trump administration's public AI strategy still leans toward speed. AI.gov describes the Action Plan around three pillars: accelerating innovation, building AI infrastructure and leading international diplomacy and security. The message is direct. America wants the largest AI ecosystem, and it wants the standards and economic benefits that come with it.
Yet the June executive order shows the administration is also becoming more alert to risk. It calls for an AI cybersecurity clearinghouse and encourages collaboration with operators of critical infrastructure such as hospitals, banks and utilities. That is not the language of a government that believes private deployment alone will solve the problem.
The pressure point is cyber capability. Advanced models that can find software flaws can also help attackers exploit them. That does not mean every frontier model is dangerous in the same way. It does mean the government is beginning to treat model release as something with national security consequences, not just as a product launch cycle.
This is where the debate inside Trump's coalition becomes important. Silicon Valley advisers want faster deployment, lighter rules and more infrastructure. Other officials are looking at cyber risk, jobs, state authority and the possibility that a model released too widely could create problems no one can quickly contain. The 30-day review compromise sits between those camps. It gives government a preview, but not enough time to turn review into a slow approval gate.
Founders should read that carefully. The near-term opportunity is still large for companies building AI infrastructure, security tools, federal software and data center services. But the idea that Washington will stay completely hands-off is fading. If your product touches frontier models, critical infrastructure, public sector procurement or automated cyber work, policy is now part of the product environment.
Krishnan's departure may make that environment more complicated, not less. A formal adviser leaving government can reduce direct authority, but an outside institution can widen reach by convening industry, publishing frameworks and shaping the language policymakers use. The next thing to watch is whether his new effort becomes a transparent forum for technical policy or a quieter place where the industry's preferred rules are drafted before the public ever sees them.
AI founders do not need to treat Washington as the center of their company. They do need to understand that Washington is becoming part of the market. The companies that notice early will be better prepared when model review, data center permitting and federal procurement rules stop being background noise and start affecting revenue, release schedules and investor confidence.
Also read: China tells fund managers to fund innovation without chasing hype • Zinc oxide and tellurium point to simpler AI chips • Musk's Terafab push puts ASML at the center of the AI chip race
