Claude.ai is a wonderful product. It makes me dramatically more productive, enhancing the quality of my work and that of my colleagues (especially, but not exclusively, at my neuromorphic concurrent compute software company).

It sure is important that there be guardrails to prevent AI from turning into Skynet and… Hasta la vista, baby!

But … whose guardrails?

Take it from an old Washington hand who has now largely migrated to the digital technology sector, yet who once served as a career civil servant in the US Department of Energy, a deputy general counsel in the Reagan White House and as senior counselor to Buzz Aldrin in his work with the National Space Council in the first Trump White House (and more).

Anthropic CEO Dario and President Daniela, you are babes in the DC woods.  Or as reported by Politico, “You can’t tell everyone that your product might destroy the world and then not expect the government to be involved,” an administration official said. “They’re politically naive.”

Anthropic’s political naivete slip showed again recently in its June 10th letter to Sens. Elizabeth Warren and Tim Scott.  Per the BBC, accusing “operators linked to Alibaba carried out almost 29 million exchanges with Claude using thousands of fraudulent accounts in what it called the largest extraction campaign of its kind…. And urging Congress to penalise the companies behind attacks like this and to ramp up measures to prevent US tech from being stolen.”

Sounds like Anthropic has a legitimate grievance. The naivete is in the remedy Anthropic proposes: export controls.

Pandora?  Don’t open that box.

Per arstechnica.com Anthropic seeks “more export controls on chips to hamstring Chinese access to advanced compute so that they simply can’t train on US model outputs.”

A cure worse than the disease.

George Gilder, author of the million-seller Wealth and Poverty, “the bible of Reaganomics,” and one of the shrewdest thinkers about technology and technology policy alive today, stated in the Wall Street Journal:

“By cutting off the Chinese chip market, which contains the majority of semiconductor engineers, U.S. industrial policies have hampered American producers of wafer-fabrication equipment—essential for making chips—without slowing China’s ascent. In the wake of these protectionist policies, launched around 2020, Chinese semiconductor capital equipment production has risen by 30% to 40% annually, compared with annual growth of about 10% in the U.S.”

Gilder makes three decisive criticisms of export controls.

They are futile, sparking foreign innovation. “The success of DeepSeek, the Chinese rival to American goliaths with radically more cost-effective artificial intelligence, reveals the futility of U.S. sanctions policies. … The chief obstacle to the success of such [cutting edge tech] ventures is the U.S. national-security apparatus, which somehow imagines that by inflicting sanctions on China, it can help Americans. …Technology is the key adventure of human progress, and it is intrinsically global.”

They are national security folly.  “Under the Biden administration, the American government was captured by some of the world’s most ham-handed national-security socialists, while the Chinese private sector under Xi Jinping commands some of the world’s most nimble capitalists.” Really, Anthropic? Back to the worst of Bidenomics?  And…

The microchip era is about to end.  “The post-microchip era, with data centers in a box of wafer scale processors, is coming. America, not China, should lead the way.”

Export controls are the Maginot Line of the 21st century. They are inherently futile and lull America into a false sense of security rather than spurring us to innovate and lead. Bad code.

Anthropic’s pattern of promoting federal regulation to achieve technological security leads with its chin, providing a veneer of legitimacy for the Secretary of Defense to declare it a supply-chain risk, attempting to ban it from the federal sector.

The Secretary made it clear that he landed on you to show that he, not Anthropic, was, no pun intended, calling the shots. He, explicitly, was not motivated by a desire to keep Claude from going rogue and destroying the world.

You deliver enormous value to various federal agencies and thus were not so easily ostracized, as it happened.  A story for another day.

Alas, that limited trauma did not immediately wise you up to the moral hazard you were cozying up to.  Anthropic obnoxiously:

1. Supports licensing requirements for AI, thereby throttling the climate that fosters innovation. Hey! If government required websites to be licensed there would be, maybe, a few tens of thousands of websites rather than the estimated 1.43 billion, of which approximately 200 million are active, serving around 6 billion people, most of whom use it daily.

Who regulates websites? The Islamic Republic of Iran. The Russian Federation. The People’s Republic of China (with its notorious “Great Firewall.”) Few Americans accordingly fine the concept of “licensing” digital technology to be consistent with liberty or prosperity.

2.     Calls for mandatory government testing and safety reviews.

By whom?

For reasons addressed below, government testing and safety review assumes discernment and agility on the part of government regulators which is not in evidence. It negates the effectiveness of the rule of law: the real power of liability for negligent or malicious behavior.

Let’s not, incidentally or accidentally, kill the magic. As Ogilvy’s Vice Chairman Rory Sutherland trenchantly observes: “When you demand logic, you pay a hidden price: you destroy magic. And the modern world, oversupplied as it is with economists, technocrats, managers, analysts, spreadsheet-tweakers and algorithm designers, is becoming more and more difficult place to practice magic -- or even to experiment with it. In what follows, I hope to remind everyone that magic should have a place in our lives -- it is never too late to discover your inner alchemist.” -- from Alchemy: The Dark Art and Curious Science of Creating Magic in Brands, Business, and Life.

3. Supports disclosure laws. This of course flies in the face of an essential component of tech business success: proprietary information. Such as Anthropic itself appropriately relies upon, for instance:

• Training Data Sources
• Model Weights and Architecture
• Security Protocols & Infrastructure.

4. Supports liability laws to sharpen the teeth of tort lawyers.  Just as I (a member of the Bar of the State of New York) would oppose proving legal impunity to anyone, AI included, I adamantly oppose  putting a litigation sword-of-Damocles over the heads of the innovators of the most vital and valuable technology since electricity, maybe fire.  It would be grotesque to treat AI like Big Tobacco or Fentanyl.

5. Opposes federal preemption of state AI regulation.  State regulation would, and probably is meant to, balkanize the marketplace, crippling the ability of startups—your potential competitors (for instance in supplanting Large Language Models with World Models)—to flourish.

6. Calls for an outright pause on global development of AI. How, pray tell, will you enforce this upon The People’s Republic of China?  Send in the United Nations Blue Helmet Peacekeeping forces?

   Pshaw! Virtue signaling at its finest!

7. Most ironically, advocates for export controls on Nvidia chips (on which, no coincidence, Anthropic’s competitors rely).  Law of unintended consequences: Anthropic got slammed with “export controls” that suppressed its most sophisticated frontier models.

Now, Anthropic recently found itself having its most advanced versions, Fable 5 and Mythos 5, abruptly, and without reasonable explanation or compelling justification, shut down by the Secretary of Commerce.

There’s an old, shrewd saying in Washington, one frequently employed by the great House Speaker Sam Rayburn:  “There’s no education in the second kick of a mule.”

The banning of Fable 5 and Mythos 5 was that second kick.

Which the president, just as arbitrarily -- after you, Dario, smartly cozied up with him recently at Évian-les-Bains -- walked back.

Maybe.

Feeling the whiplash yet?

The burgeoning multi-trillion-dollar artificial intelligence sector dramatically reshapes the frontiers of modern technology.

Yes, this must be done responsibly. No Skynets!

Who should be making policy?  Governmental personnel are not infallible. Let’s not conflate them with the Pope speaking ex-cathedra.

For structural reasons beyond their control, political appointees and career civil servants simply cannot be as discerning, agile, or accountable as tech’s front-line figures.  Here are some of the reasons why.

The judgment of Team Government is not improved by the fact that most political appointees are primarily selected for political loyalty and palatability. Not expertise.

Career civil servants (which for a thousand days I was, including my stint in the White House) are burdened with cumbersome red tape.  Such as the 5000-page regulation that must be complied with to purchase anything over $15,000.  (Its defense department counterpart regulation apparently has not reliably thwarted the sale of $600 toilet seats to the military.  D’oh!)

And consider: the merit system exams for civil service employment were suspended decades ago.  And never restored.

Meanwhile, it is prohibitively hard for Senior Executive Service managers to fire – or even give a less than stellar performance review despite manifest performance mediocrity to –  the career civil servants they are charged with managing.

Not a recipe for excellence-in-regulation.

Meanwhile, sub-cabinet political appointees mostly get rotated in, out and sideways so often that few fully master their agencies… even if they have subject matter expertise.

Moreover, the civil service system tends to punish innocent failure harshly, while rewarding major successes tepidly.

That leads to a culture of timidity.

Timidity is not the same as, or a recipe for, security.

Nvidia, led by Jensen Huang (who, had he not be born in Taiwan and thus Constitutionally barred, would be my preferred candidate for president in 2028) views AI as a tool to expand human capacity. He has proven himself a practical visionary.

He is an unequivocal social and economic benefactor who has contributed no less than $5 trillion-with-a-t in value to the American economic base.

Really, more.

Not even Elon Musk comes close to rivaling that.

Huang’s ethos, embedded in Nvidia, aligns with those of the Nobel laureate Professor Paul Romer’s breakthrough work which proves that ideas, knowledge, and technological discovery are the primary engines of prosperity and the attendant human flourishing.

Along with the late Nobelists Gary Becker and Ted Schulz’s work on human capital.

And Nobel laureate Friedrich Hayek famously showed that central planners lack the localized, distributed knowledge needed to manage complex systems.

There are reasons that the infamous central planning of the Soviet Union caused the economy of the USSR to stagnate and then, collapse.  Let’s not emulate nor echo the failed Soviet model!

Anthropic?  Do not become the tool of the Establishment. The push for heavy regulation ultimately threatens to privilege those with political motives together with the plodding, risk-averse bureaucracy (in cahoots with entrenched federal prime contractors), over an innovative free market, prejudicing disruptive startups by saddling the sector with untenable compliance costs.

Anthropic?  Love your product!  That said, stop being politically naïve.

Anthropic? You naively thought federal regulation would create security.

Instead, it weaponized the Establishment (prompted by one of your main competitors) … against you.

That is not the road to technological security.  The road to Hell is paved by intentions as good as yours.

Your intentions are noble.

Your proposed policies?

Not so much.

Anthropic? For your own sake, for America, and for the world:

Do not court a third kick of the mule.