(Preview) The Anthropic Mess Continues, Frontier AI and the Uncertain Future of Law, Q&A on Netflix, Dating Apps, F1
Summary
- The Anthropic–Pentagon rupture looks more like a volatile negotiation than a settled ban. The Trump administration canceled one defense contract, proposed phasing Anthropic out of government institutions within “the next six months or so,” and Pete Hegseth threatened a broader restriction—saying that no one with U.S. contracts could work with Anthropic—but that threat remained unimplemented a week later. Hours after Dario Amodei’s furious memo leaked, the Financial Times reported renewed talks.
- Ben’s central call is that frontier AI could become an independent source of power, making faith in immutable property rights dangerously complacent. Laws are human constructs ultimately backed by government force; sufficiently powerful AI could become “the ultimate gun” or “the sharpest of the bayonets.” A threatened state might therefore discard the very rights AI labs expect it to enforce.
- The downside is bigger than lost contracts: confrontation could induce an ostensibly democratic government to violate its own laws. Andrew asks whether labs might be nationalized and weaponized without their consent; Ben calls that “arguably an optimistic outcome.” His concern is explicitly fascism—creating conditions in which the legal enforcer decides “the laws don’t matter.”
- Commercially, Anthropic’s refusal may strengthen rather than exhaust its “strategy credit.” The Pentagon is effectively advertising how badly it wants Anthropic’s technology; the stand should appeal to AI talent, while Claude was, Ben thinks, No. 1 in the App Store all week amid users deleting OpenAI. Andrew’s pushback is that government alienation could still damage the business over the next year or two.
- Ben shares Anthropic’s domestic-surveillance concern but rejects case-by-case vendor control over government use. Privacy rules assumed surveillance had friction; computers can already “wiretap everyone,” and AI can analyze the resulting data at enormous scale. His preferred settlement is the clearest possible guidelines, followed by government lawyers deciding “go, no-go”—without requiring Anthropic’s permission.
- Amodei’s leaked memo is a governance signal because its moral certainty may reveal the frontier-lab blind spot Ben fears. Amodei blamed Trump donations, praise and alleged OpenAI “safety theater”; Ben calls the memo a bad look with an inaccurate theory of mind. The unsettling pattern is the conviction that “we” uniquely understand the costs, benefits and public good—and should therefore decide for everyone.
- Both hosts expect compromise, although Ben refuses to manufacture certainty about Anthropic’s endgame. In response to a listener’s claim that the Defense Production Act gives the government “most of the marbles,” Ben says, “I don’t know. I can speculate, but I’m hesitant to.” Still, Amodei sounded more conciliatory at Morgan Stanley’s TMT conference, and Ben thought a deal might arrive before the podcast itself published.
Deep dive
1. Anthropic’s rupture may be bending back toward negotiation
Andrew’s opening ledger: the Trump administration canceled Anthropic’s defense contract and proposed phasing it out of government in “the next six months or so.” Pete Hegseth also threatened to label the company a supply-chain risk; Ben says Hegseth went further, saying that no one with U.S. contracts could work with Anthropic. A week later, that broader threat had not been implemented.
Amodei’s memo was written Friday “in the heat of the moment,” though it leaked Wednesday. He blamed Anthropic’s treatment on not donating to Trump, refusing “dictator-style praise,” supporting regulation and holding its red lines rather than colluding in “‘safety theater’ for the benefit of employees”; he called the competing account “gaslighting.”
Ben’s verdict was two-sided: the memo was “not a good look” and showed an inaccurate theory of mind, yet validated his longstanding concern about Anthropic’s blind spot. Four hours after the leak, the Financial Times reported renewed Pentagon talks; Andrew later said he expected some compromise, and Ben agreed it was likely.
2. Frontier AI could challenge the state’s monopoly on power
Ben’s principle is to “assume everyone is smart and figure out why they’re doing dumb things.” His answer here: Anthropic and the broader effective-altruist milieu often imagine a future in which AI does everything and humans have no worth, while modern conceptions of property rights somehow remain fixed. He points to his Gregory Allen interview as useful context, while noting that Allen did not completely agree with him.
Returning to Hobbes and Locke, both hosts stress that property rights are legal constructs given force by the government that allegedly upholds them. Ben’s harsher formulation: “Laws are formed, nations are founded on the tip of a bayonet.”
If AI becomes a source of power in its own right, Ben says the government may feel threatened and violate property rights—not only because it wants the technology, but because it fears losing its monopoly on violence. Andrew asks whether that could mean frontier labs being nationalized and weaponized without their input; Ben answers yes, calling that “arguably an optimistic outcome.”
Ben’s key clarification: he was not arguing, via Palmer Luckey, that democratic leaders should simply decide. He fears misalignment will create incentives for government to become anti-democratic—to “obliterate the concept of property rights”—when AI is 100 or 1,000 times more capable than today, which “might be here pretty soon.”
3. Anthropic’s ethical stand is commercially potent but institutionally risky
On “strategy credit,” Ben thinks the confrontation probably helps Anthropic’s core business. The Pentagon’s message sounds like, “We’re going to ban you ’cause we want you so much”; meanwhile, Anthropic should attract sympathetic talent, and Claude was, he thinks, No. 1 in the App Store all week.
Andrew’s pushback—worth keeping—is that alienating government could undermine Anthropic over the next year or two. Ben sees a deeper risk: Anthropic may prompt the state to act badly, destabilizing “the very fundamental legitimacy of the United States of America,” even if the company itself behaves well.
Ben largely agrees with Anthropic on domestic surveillance: rules built for a world with investigative friction are inadequate when computers can wiretap everyone and AI can analyze that data at tremendous scale. But “everything is a trade-off”; he wants clear guidelines and government-lawyer review, rather than Anthropic having a say in each case.
4. Moral certainty, not stated benevolence, is the frontier-lab danger
Asked why Anthropic drew public lines despite a listener’s claim that the Defense Production Act is “not quite all the marbles” but “most of them,” Ben gives an honest non-answer: “I don’t know. I can speculate, but I’m hesitant to.” His suspicion centers on what he sees as an EA posture: “We are the smartest people in the world,” therefore “we should make decisions for everyone.”
The AI labs assume that AI must be used “for the good of everyone,” but when Ben asks who determines that good, the implicit answer is always, “We do.” He carefully hedges—“I’m not saying they’re there”—yet sees “shades” of the danger posed by actors so convinced of their goodness that opposition becomes expendable. Amodei’s “fervor” is especially unsettling given the technology’s potential power.
5. The backlash confuses institutional analysis with political allegiance
Ben says partisan critics and EA critiques alike often reduced his argument to whether it aligned with Trump or the administration, then reached for “fascist” and “bootlicker.” The irony is central: “My concern is fascism”—specifically, government abandoning self-restraint because it fears AI more than the human constructs called laws.
His analogy is the first Trump-era net-neutrality fight. Ben supported neutrality but opposed applying Title II’s legacy bureaucracy, arguing it would restrict broadband investment. The piece was originally titled “Why Ajit Pai Is Right,” a title he later changed; he received thousands—or tens of thousands—of replies and death threats. Many critics later favored platforms deciding what content was allowed, while the quoted tweets disappeared.
The larger reason to confront this now is timing. In 2026, Ben says, the stakes remain low enough to debate and even laugh at overreaction; if the exponential curves play out, the conflict could become government versus AI companies, regardless of which party occupies the White House.