Anthropic Isn't the End of Man

“The world is rightly aflutter this week over the "PHP" vulnerability, one of many cyber bugs exposed by AI that the world's cyberwarriors are desperately trying to fix one step ahead of the world's hackers. But notice this is already different from the science-fiction fears birthed before AI was a real thing, when simple premises permitted so many startling conclusions.

 

Remember when artificial superintelligence would outsmart every human effort to control it, because, see, it would be superintelligent? There would be "foom" -- it would explode on the world overnight and have us under its thumb before we knew it. This would happen when AI starts training AI, known as "recursive self-improvement."

 

But then reality happens. Fears that once seemed realistic stop seeming realistic. AI companies, it turns out, don't start an automated training process and come back later to see what it hatched. Any fooming process would quickly be self-liquidating, exhausting its supply of chips and power.

 

OpenAI CEO Sam Altman perhaps doesn't control the human future after all since he doesn't even control his own company, owning zero percent of the stock. Or its flagship chatbot, which will happily tell you Mr. Altman is "increasingly questioned as a trustworthy steward of a technology that could reshape society."

 

Now comes the most awaited stock offering of the AI era. Missing the point, Bloomberg headlines its story: "Anthropic Flags 'Existential Risks' in Long-Awaited IPO Filing."

 

Uh huh. Put it this way. Buying the stock, you might lose money. But if extinction is humanity's fate, not buying it won't save you.

 

Happily, real-world facts are starting to influence the story arc. Martin Peers of The Information points out that "Anthropic's ability to even release its best models is diminishing" due to government intervention.

 

"After Mythos" -- the ultra-cyber-capable model Anthropic felt obliged to withhold from the market in April -- "the whole world changed in D.C.," says anti-AI organizer Connor Leahy.

 

Aleksander Madry, an MIT professor, unexpectedly returned to a recently abandoned job at OpenAI to participate in what he calls a "different regime of AI progress" centered on safety and security challenges.

 

The real gist of Anthropic's warning to investors? They are entering terra incognita in how society will treat and regulate this new technology -- though don't worry, those who warned of human extinction will soon be saying "I told you so" when AI mistakenly deletes your email.

 

Electricity, to which AI is often compared, came into the world before the last century's giant expansion of government and the national-security state. More significantly, AI landed at a moment when cyber was a mission-critical focus of the U.S. defense and national intelligence establishment. That's practically all you need to know. The extent of this importance is hard to exaggerate -- from monitoring the world's vast and growing data flows to protecting the data and infrastructure on which the fully digitized U.S. economy now relies.

 

Anthropic's prospectus mentions "existential risks to humanity" three times. It mentions national security 22 times. If coming models are as powerful as their antics in the lab indicate, their availability to the public, if any, will be strictly controlled by Washington. If you have doubts, look at the stock market. No obvious end-of-the-world trade is happening, but shares of cybersecurity companies have been straight up since Mythos was unveiled.

 

This week economist and longtime AI cheerleader Tyler Cowen conceded the obvious to the New York Times. The prevailing political dynamic "puts more weight on our national-security establishment and the Defense Department. I don't feel entirely comfortable with that, but I think that's what it's going to have to be."

 

Yep. Don't expect to get your hands on the kind of ultra-sophisticated tools that will be available only to Anthropic's "trusted-access" partners -- i.e., those approved by government. The midwifing of advanced AI into the world economy will increasingly be controlled by two bureaucracies, ours and China's.

 

But a separate phenomenon remains the availability of increasingly sophisticated chatbots, acting as adviser rather than agent. These are less likely to be inhibited by government, partly for First Amendment reasons in the U.S. This might be a bigger revolution than we appreciate.

 

Why? Humans are highly social apes with big brains -- and these things have eternally been in tension. You might have learned as much from the allegory of the emperor's new clothes or the discoveries of behavioral economics. The progress may not be dramatic. It might even be subliminal. But a large and unexplored scope exists for improving society's collective thinking processes. We obviously need help. That help could be arriving just as democracy confronts daunting demographic, fiscal, national-security and -- yes -- technological challenges.” [1]

 

1. Anthropic Isn't the End of Man. Jenkins, Holman W; Jr.  Wall Street Journal, Eastern edition; New York, N.Y.. 28 Oct 2026: A15.