This month, a single letter took a frontier AI product to zero in about six hours. No court, no notice-and-comment period, no phased rollback. Commerce Secretary Howard Lutnick sent Anthropic an export-control directive on a Friday evening, and by midnight the company had disabled Claude Fable 5 and Mythos 5 worldwide for every customer to comply. Three days earlier, Fable had launched to real fanfare. I am going to leave the question of who was right where the public record leaves it, which is unresolved, because the lesson does not depend on the answer.
Here is the thesis. Regulatory exposure is no longer a compliance function you bolt on after product-market fit. For anyone building in frontier technology, dual-use, defense, or anything adjacent to national security, the government is now a stakeholder with a hand on your off switch. That is a feature of your architecture, not a footnote to it. The companies that internalize this early will build differently, raise differently, and survive shocks that take their competitors offline overnight.
The Off Switch Was Always There. You Just Could Not See It.
The instinct when a story like this breaks is to treat it as a one-off. A specific company, a specific capability, a specific weekend. That instinct is wrong, and it is expensive. What the Fable episode exposes is a structural truth that was always latent and is now operational: a government letter, sent with no published technical rationale, can move faster than your legal team can read it.
For most of the last two decades I have watched founders model their risk around the things they could see on a cap table or a competitive landscape. Will a bigger player copy us. Will the round come together. Will the key engineer stay. Those are real risks. But the most consequential variable for a whole class of companies now sits outside the building entirely, in a regulatory apparatus that can act in an evening. And the cruelest part is that the capability that makes your product valuable, in this case the ability to find and explain software vulnerabilities, is frequently the same capability that makes a national-security stakeholder nervous. That is not a bug in your business. For dual-use companies, it is the business.
Build the Switch Into the Blueprint
If the government can turn your product off, the engineering question becomes: what survives when it does? This is where the difference between a fragile company and a resilient one gets decided long before any letter arrives.
The strongest posture I have seen is architectural separation between the capability that creates value and the capability that creates regulatory exposure. Anthropic actually understood this. Fable shipped with classifiers built specifically to suppress the high-risk capabilities of the underlying model, and the company reports red-teaming those safeguards for thousands of hours with government and outside institutions before release. That is more diligence than most startups will ever do. And it still did not prevent a shutdown, which tells you something important: doing the safety work is necessary, but it does not buy you control over how a stakeholder interprets it under pressure.
So the blueprint has two layers. The first is technical containment you can demonstrate, not just describe, because in a fast-moving dispute the burden of proof lands on you and verbal assurances do not travel well. The second is operational continuity that assumes the high-exposure feature can be severed without killing the company. If your entire revenue line depends on the one capability most likely to draw a directive, you have built a company with a single point of political failure. The founders who think about this early build modularity that lets them keep serving customers in a degraded but functional state while a dispute plays out. The ones who do not, go dark.
Your Safety Story Is a Promise You Will Be Held To
There is a second-order lesson here that is uncomfortable and worth sitting with. The more loudly you position yourself as the responsible, safety-first operator in your category, the more your conduct in a crisis gets measured against that standard rather than against your competitors. Part of the criticism leveled at Anthropic was not about the technical facts at all. It was that a company whose entire brand is built on safety appeared, to some observers, to resist a safety-motivated request. Whether that characterization is fair is genuinely disputed. But the dynamic is real and it generalizes.
When you tell the government, your customers, and the market that you are the disciplined, mission-aligned choice, you are issuing a promise. That promise earns you trust, access, and often a seat at tables your competitors cannot reach. It also creates a gap that any adverse moment will be measured against, and stakeholders with power will hold you to your own stated standard more strictly than they hold the company that never made the claim. For Moonshots, where roughly a third of our capital backs veteran-founded teams and a third sits in dual-use technology, this is not abstract. The founders we back lead with mission and integrity because it is true to who they are. The discipline is making sure the operational reality can always back the positioning, because the day it cannot is the day the positioning becomes a liability.
Sovereignty Is Now a Sales Objection
Watch what happened within 24 hours of the shutdown. A Chinese lab launched a competing model and built its pitch around the unreliability of American AI as an export partner. That is the part of this story getting the least attention and deserving the most, at least if you sell frontier technology across borders.
When your government can unilaterally cut off your foreign customers, those customers learn the lesson faster than you do. They start asking about sovereignty, about on-premise deployment, about whether they are building on infrastructure that a letter from Washington can disable. The demand for sovereign AI does not disappear when this particular dispute resolves. It compounds. If you are building anything that crosses borders, the question of who can switch you off is now a live sales objection, and the answer “trust us, that won’t happen” stopped being credible the moment that letter went out.
What Extraordinary Founders Do With This
The reflexive response to a story like this is fear, or a retreat from dual-use altogether. That is the wrong read, and it is the read that cedes the most important categories of the next decade to whoever has the stomach to stay. Defense modernization, AI infrastructure, identity, space, and logistics resilience are not getting less entangled with government. They are getting more entangled, which means the regulatory surface is not a reason to avoid these markets. It is the terrain you have to learn to operate on.
The extraordinary founders I back will do three things with this episode. They will treat regulatory relationships as a core competency built before they need it, not a fire drill run during a crisis, which means knowing the people in the relevant agencies by name and earning credibility in the calm before there is ever a dispute. They will architect their products so the highest-exposure capability can be contained, demonstrated, and if necessary severed, without taking the whole company to zero. And they will make sure their safety positioning describes how they actually operate, because the gap between the story and the substance is exactly where a Friday-evening letter does its damage.
A government letter took a frontier model offline in six hours this month. That is the new physics of building at the edge of what technology can do. The leaders who internalize it will not be the ones who avoided the hard markets. They will be the ones who built for the off switch before anyone reached for it, and were still standing when the dispute was over.
Kelly Perdew is the Managing General Partner of Moonshots Capital, an early-stage venture firm he co-founded with Craig Cummings. He has been investing in startups since 2004 and has backed 131 companies with 33 exits and 19 unicorns including LinkedIn, Pandora, ID.me, and Scopely. Moonshots Capital invests in extraordinary leaders building transformative technology.
