The Last Technical Voice Leaves the Room
In the late spring of 2024, a mid-level cybersecurity analyst at a federal agency sat in a windowless conference room in Virginia, watching a demonstration. On the screen, an AI model found a vulnerability in a piece of critical infrastructure software that a team of human experts had spent three months trying to locate. The analyst told a colleague afterward that what unsettled him was not the speed — it was the silence. [1] The machine did not celebrate, did not pause to consider the implications, did not ask whether the flaw should be reported or weaponized. It simply found the next hole and moved on. That moment, he later said, was when he understood that the rules of national security had changed in a way his agency was not ready to handle.
Thomas Lind understood that moment too. As the head of policy at the Office of the National Cyber Director and a senior adviser to National Cyber Director Sean Cairncross, Lind was one of the few people inside the White House who could translate what that AI model had done into language that policymakers could act on. He came from the National Security Agency, where he had spent years watching how adversaries probe digital defenses. He knew that the gap between a theoretical risk and a deployed exploit had shrunk from months to days, and that frontier models from companies like Anthropic and OpenAI were now capable of finding software flaws faster than any human team could patch them. That was the problem he was hired to solve.
Now he is leaving. Lind plans to exit government soon to spend more time with family, according to two people familiar with his decision. His departure follows that of Alexandra Seymour, his former deputy on policy, who left last week. Sriram Krishnan, the White House’s senior policy adviser for AI, plans to leave later this month. The exits come just days after the release of the long-awaited executive order on AI security — an order that had been delayed repeatedly due to infighting over how heavily to regulate the technology. The timing is not coincidental. It is the sound of technical expertise walking out the door just as the government needs it most.
To understand why this matters, you have to understand what the Office of the National Cyber Director was supposed to be. Congress created the ONCD in 2021 to oversee the budgets and cyber defense policies of the sprawling federal executive. The idea was that the federal government needed a single office that could see the whole picture — not just the threats from state-backed hackers or criminal syndicates, but the emerging risks from technologies that were evolving faster than the bureaucracy could track. The national cyber director was supposed to serve as the principal adviser to the president on everything from deterring malicious cyber activity to shoring up U.S. cyber defenses. It was a mandate that required deep technical understanding.
But the office has never been fully staffed. Cairncross has filled only about three dozen of the 75 authorized positions, according to reports. His chief of staff and deputy chief of staff lack tech or cybersecurity experience. Cairncross himself is a former lawyer and Republican National Committee executive who had no background in cybersecurity or digital policy before his confirmation in August 2025. In a field where technical credibility is the currency of trust, the ONCD has been running on a thin reserve. Lind was one of the few people who could spend that currency effectively.
The executive order that just came out lays out actions the ONCD and other federal agencies must take in the coming weeks to fortify networks against the hacking risks posed by frontier models. It is a document that assumes the government has people who understand what those models can do — not just in theory, but in practice. Lind helped craft that order. He participated in meetings with Google, Anthropic, and OpenAI as they provided feedback. [2] He knew which questions to ask and which assumptions to challenge. Without him, the implementation of that order loses its most technically literate interpreter.
The executive order’s delays stem from a fundamental mismatch: the technology is evolving faster than the government’s ability to hire and retain people who understand it. Whether those criticisms are fair or not, they point to a deeper problem: the White House is trying to regulate a technology that is changing faster than its personnel can adapt. The people who understand the technology are leaving, and the people who stay do not always understand what they are regulating.
This is not a new problem for the federal government. The gap between technical expertise and policy authority has been widening for decades. During the early days of the internet, the government outsourced most technical decisions to engineers and academics who understood the infrastructure. As the internet became central to national security, that model broke down. Policymakers started making decisions about encryption, surveillance, and network security without fully understanding the trade-offs. The result was a series of policy failures that ranged from the merely wasteful to the genuinely dangerous.
Artificial intelligence is repeating that pattern, but at a much higher speed. The models that can find software vulnerabilities are not hypothetical. Anthropic announced in April that it had built a model so adept at finding holes in digital infrastructure that it could enable criminals and state-backed operatives to launch a barrage of cyberattacks if made available to the general public. [1] That announcement kicked the White House’s effort to craft new guardrails into high gear. But the people who know how to evaluate such claims — who can distinguish between a genuine breakthrough and a marketing exaggeration — are becoming scarce in the offices that need them most.
The ONCD is actively seeking qualified candidates, but the exodus of technical talent has left the office struggling to evaluate the very AI models it is tasked with regulating But hiring technical talent in a competitive market is difficult even when the government offers competitive salaries and clear career paths. It is nearly impossible when the office has a reputation for low morale and internal dysfunction. Several people familiar with the ONCD’s internal dynamics said others from the White House cyber policy shop are looking to exit due to low morale. The talent drain is not a leak — it is a structural failure.
While White House spokesperson Liz Huston defended Cairncross’s leadership, the reality is that no amount of praise can replace the technical expertise needed to understand how AI models are finding vulnerabilities faster than humans can patch them. It does not find the vulnerability that a frontier model can exploit in seconds. It does not replace the institutional memory that walks out the door when someone like Thomas Lind decides that the cost of staying has become too high.
Sriram Krishnan, who is leaving at the end of June, plans to launch an outside institution that will still give him a role in influencing AI policy. In his departure post, he wrote that it had been an honor to serve under President Trump and credited outgoing AI and crypto czar David Sacks as his closest collaborator. Sacks called Krishnan’s skills genuinely unique and said his departure would be a huge loss for the administration. The irony is that the administration is losing the very people it needs to implement the policies it just announced.
The pattern is not limited to this administration. Every White House in the last decade has struggled to retain technical talent. The private sector offers higher pay, more autonomy, and the satisfaction of building things rather than navigating bureaucracy. But the stakes are higher now. The models that can find software vulnerabilities are not science fiction. They are being deployed by companies that are accountable to shareholders, not to the public. The government’s ability to understand and regulate these models depends on having people who can ask the right questions. When those people leave, the government becomes a consumer of information rather than a producer of understanding.
Lind’s departure is the latest in a string of exits from the White House’s tech policy ranks since the release of the executive order. It is not the first and it will not be the last. The question is whether the institutions that remain can absorb the loss and continue to function. The ONCD was designed to be the nerve center of the federal government’s cyber defense. A nerve center without nerves cannot react. It can only report what has already happened.
The loss of technical expertise in the White House means that the government is increasingly unable to distinguish between genuine AI breakthroughs and marketing exaggerations, leaving critical infrastructure vulnerabilities unaddressed.
Sources
1. Anthropic
2. Google