The timing of Sam Altman's principled stand is a bit suspicious

[Standard caveats for the Hugging Face hacking thread.]

[The mainstream coverage of the story has been sensationalistic and alarmist, with many details badly reported or out-and-out wrong. That doesn't mean you shouldn't be scared. LLM-based agents are getting disturbingly good at breaking through cybersecurity, and in this one area, they will only get better.]

[The potential for extensive and sometimes catastrophic harm is immense. You don't have to believe in Rogue AI and the coming robot apocalypse. If anything, some of the skeptics, like Cal Newport and Gary Marcus, are among the most concerned.]

[Imagine a world where every bad actor, from cyberterrorist all the way down to two-bit scam artist, had access to the world's greatest team of hackers. Or a world where any moron playing around with a computer could accidentally shut down the power grid. You don't need stories of Rogue AI to make that scenario scary.]

I've seen a lot of True Believers online railing against the argument that the uproar over the Hugging Face incident is all hype being generated by the frontier model companies for their own nefarious reasons. Curiously, though I follow the skeptic community (Marcus, Newport, Carl Brown), I've never actually come across anyone making these charges. The consensus in that group seems to be that these safety concerns are both real and disturbing. Gary Marcus has gone so far as to describe them as potentially catastrophic.

That said, it's important to make the distinction between "all a massive conspiracy" and suggesting that some of the executives at these companies might be trying to spin recent events in their favor. There are literally trillions of dollars at stake here in an industry that was widely considered a bubble of unprecedented size.

Which brings us to...

OpenAI delays IPO as Sam Altman cites growing safety concerns
Aidin Vaziri

OpenAI will not go public this year, CEO Sam Altman said in an interview released Saturday, delaying one of the most anticipated stock market debuts in Silicon Valley as the artificial intelligence industry faces growing scrutiny over safety.

Altman told Fortune that taking the San Francisco company public now would be the wrong move while OpenAI and other major developers work through questions about how quickly to build more powerful systems.

This story has been getting a lot of play. Based on the headlines and ledes, choosing to forgo a trillion-dollar IPO out of safety concerns would certainly indicate that Sam Altman and OpenAI were taking these concerns very seriously.

If it were true.

If you've been closely following the story, or just reading below the fold in most of these articles, you would find that there has been extensive, well-sourced reporting from credible sources showing that the decision to delay the IPO until 2027 was made a couple of months ago for reasons that had nothing to do with safety.

Here's Forbes from late June:

OpenAI hired bankers and lawyers eyeing an initial public offering as early as the third or fourth quarter of this year, with CEO Sam Altman pushing them to engineer a $1 trillion valuation, the Times reported, citing three people involved in the talks.

Over the past week, OpenAI’s advisers have cautioned the company that a public listing may not be met with enough enthusiasm due to the volatile public tech market, per the report.

When advisers offered a choice between waiting until 2027 for a $1 trillion debut or accepting a lower valuation for a faster one, Altman called any cut to the trillion-dollar figure a "nonstarter," one person in contact with him told the Times.

At the risk of putting too fine a point on it, being able to rebrand the delay from from being a sign of a struggling company to being a bold statement of principle is a big win for Altman and his company. The original story had been a huge black eye and had come at a time when a growing number of analysts and commentators were speculating that OpenAI was struggling to survive, particularly when compared with Anthropic.

Once again, this is not meant to suggest that they did it on purpose. The PR angle alone is not enough to make up for the legal and regulatory concerns that the Hugging Face incident has likely put into motion.

There is, however, at least one plausible scenario where OpenAI's cybersecurity fuck-up could actually be the best thing to happen to the company in years. Anthropic being forced to delay its IPO due to safety concerns and the inspiring example of its competitor would be a major lifeline for OpenAI. (Though so far it doesn't seem to be working.)

Hugging Face was a major body blow to an industry that was already about as popular as a venereal disease. It's silly to assume the leaders of these companies wanted it to happen, but it's naive to assume they'll be completely honest and altruistic about their response either.
 

Sanewashing Revisited

While reading Andrew Gelman's recent post on sanewashing, I was surprised at how many people in the comment section seem to be completely unfamiliar with the term. Best lesson here is that I shouldn't assume that most normal people obsessively follow press critics like Margaret Sullivan.

With that in mind, here is my working definition, supported by a beautiful example from, of course, the New York Times.

Sanewashing is the practice of journalists downplaying signs of instability, mental decline, and general craziness through often comically understated language, extensive and misrepresentative paraphrasing that overstates the coherence and softens the rhetoric of the original statements, burying especially embarrassing details, or simply choosing not to cover certain stories.

Sanewashing is closely associated with normalization, coverage that desensitizes readers to how extreme and bizarre these times are.

Sanewashing is applied to various groups (Silicon Valley lunacy has been benefiting from it for well over two decades), but it is most strongly associated with MAGA and Donald Trump.

When on the lookout for sanewashing, it is important to note that, with the exception of stories that receive no coverage at all, the process almost never washes away all of the craziness. Though we can't get into the reporters' and editors' heads, the objective generally appears to be finding a middle ground between depicting these statements, events, and policies as normal and calling them out for what they are.

Last week's Republican "convention" gave us numerous fine examples, but this might be the best.

Trump: "Please right your right hand. I pledge to the greatest president in the history of the United States. That loves us so much he can't even breath. That I will go out with my family, my friends, I'll do it any way - I don't care if I'm registered or not, I'm going to try to cheat like hell..."

[image or embed]

— Aaron Rupar (@atrupar.com) September 10, 2026 at 7:32 PM

 

At one point, Trump, invoking the extensively debunked lie that massive voter fraud cost him the election in 2020, had Republicans pledge to do anything necessary, including cheating and voting illegally, in order to win in November. You can see the video in the tweet. Here's the transcript.

TRUMP: So please raise your right hand. I pledge —
CROWD: I pledge.
TRUMP: To the greatest president in the history of the United States —
CROWD: To the great president in the history of the United States.
TRUMP: That loves us so much he can't even breathe.
CROWD: That loves us so much he can't even breathe.
TRUMP: That I will go out with my family, my friends, I'll do it any way, I don't care if I'm registered or not, I'm going to try to cheat like hell like they do. There has never been bigger cheaters, they don't care, I am going to go out and I'm going to get my friends and family and we are going to vote on November 3, or we are going to vote before that. 

Now here's how the New York Times described that moment.

By Shane Goldmacher, Lisa Lerer Sept. 11, 2026

Vice President JD Vance was billed as the headliner of the second night of the Republican convention, but President Trump stole the show.

With a hug and a handshake, Mr. Vance exited stage right after his speech as Mr. Trump returned to the place he likes most — center stage — to make his final pitch of the convention to keep Republicans in power.

At one point Mr. Trump asked attendees to raise their right hands and pledge to him that they would vote with their family and friends. Dutifully, they did.

“You know what happens if you don’t vote?” Mr. Trump said. “You go to hell.”

 You'll notice the most memorable and newsworthy bit have somehow been left out.

Perhaps an even more dramatic example of sanewashing was found in the video that accompanied the New York Times story. I don't know how to embed this video, but I did capture the relevant frames.

The video captures the opening of Trump's statement, but abruptly stops mid-sentence right before the part about cheating.

By comparison, PBS Newshour ran the entire clip so make sure to do your part when the next pledge drive comes around. 

Through not directly related to sanewashing, this RNC video gives you some idea of what an accurate depiction of today's GOP would have to convey.

Wow -- RNC video: "On June 14th, 1946, God looked down on his planned paradise and said, 'I need a caretaker,' so God gave us Trump." It only gets more unhinged and culty from there.

[image or embed]

— Aaron Rupar (@atrupar.com) September 10, 2026 at 5:26 PM

 

Nothing untoward going on here

DRUCK: “.. Given what’s going on in the economy and the capital spending boom and the war for capital, if anything, [bond yields seem] a little low. .. It’s just been like a slow, fundamentally driven march upward in yields. @financialtimes.com www.ft.com/content/7410...

[image or embed]

— Carl Quintanilla (@carlquintanilla.bsky.social) September 10, 2026 at 11:00 AM

Damn, Stan.

[image or embed]

— Carl Quintanilla (@carlquintanilla.bsky.social) September 10, 2026 at 11:08 AM

From the FT: [with emphasis and annotation added.] 

Anthropic and OpenAI’s bankers are lobbying for an investment-grade credit rating after their upcoming initial public offerings, a designation that would lower the borrowing costs for their ambitious AI infrastructure plans.

Morgan Stanley and Goldman Sachs have held talks with credit rating agencies in recent weeks on behalf of the two leading AI labs, as they look to gain access to the $11.7tn corporate bond market post-IPO, said people familiar with the matter.

...

Achieving an investment-grade rating from Fitch, Moody’s and S&P soon after going public would be a remarkable feat for the two lossmaking AI labs, unlocking big benefits for the companies and their infrastructure partners including Oracle and Nvidia.

The rating would open the door to pension funds, insurers and other institutional investors that take far more limited positions in riskier speculative-grade debt.

It would provide another example of Wall Street changing longstanding practices to usher in the three largest IPOs in history. SpaceX, which went public in June, was the first large tech company to receive an immediate investment-grade rating.

Elon Musk’s rocket conglomerate also benefited from changes to index rules that meant billions of dollars in passive investment tracking the Nasdaq immediately flowed into its stock.

Previous tech heavyweights such as Meta, Netflix and Tesla waited a decade or more after their listings to get a top-tier credit rating.

...

 A ratings bump could also help Oracle refinance some of its current debt pile after raising funds to fulfil a $300bn data centre build-out for OpenAI that has put it at risk of losing its investment-grade status following a recent downgrade.[Because it would be terrible if anything were to happen to the Ellison's massive fortune. -- MP]

 ...

Anthropic is expected to unveil its IPO prospectus soon, allowing investors to pore over its finances ahead of a listing that could value the five-year-old company at $2tn or more. OpenAI is expected to follow suit with an IPO next year. [That 2027 date is key context for the Hugging Face incidents. The company had already been forced to push its IPO forward from 2026 and really needed a win which probably had something to do with the criminally  negligent corner-cutting which led to the hacks -- MP]

Analysts said opaque finances and the start-ups’ use of flattering annual recurring revenue figures have masked their actual performance. [And in the distance you can hear Ed Zitron laughing. -- MP]

“Both companies have not provided concrete details,” one credit analyst said.

 

 

It turned out that Microsoft CEO Satya Nadella actually could overstate the breakthrough of the metaverse

Occasionally, sitcoms will go meta and play with the conventions of the genre, such as the way that huge, life-changing events seem to disappear from the characters' collective memories and are never referred to again. For example, years after Married... with Children dropped the disastrous storyline about adopting a small child, the actor's face was briefly glimpsed on a milk carton on the kitchen table.

Both Web3 and the metaverse sometimes feel like one of those abandoned storylines. It seems like they should still be a bigger deal, the sort of thing people would bring up more often, particularly given the fact that so many still-powerful individuals and institutions, from the company formerly known as Facebook to the New York Times ("Investors Snap Up Metaverse Real Estate in a Virtual Land Boom" -- Nov 2021), invested so much of their reputations in these next big things that seem inexplicable in retrospect and which many of us were calling out in real time.

In one of his recent epic posts, Ed Zitron specifically made the connection between the concept of hypernormalization and the strange silence about the collapse of the metaverse starting with this quote from Adam Curtis’ documentary Hypernormalization:

     The Soviet Union became a society where everyone knew that what their leaders said was not real because they could see with their own eyes that the economy was falling apart.

    But everybody had to play along and pretend that it WAS real because no-one could imagine any alternative.

    One Soviet writer called it "hypernormalisation".

    You were so much a part of the system that it was impossible to see beyond it.

    The fakeness was hypernormal.

Here's Zitron: 

While both the media and world governments were adept at making moves to sustain systemic thinking — that the system or systems know best and must be protected (we must prop up banks, we must make sure we’re all in offices, etc.), big tech tried and failed twice to force society to back their concepts, the first being a VC-led attempt to make Clubhouse the next Facebook (with the media attention to match it), the second being the doomed attempt to make the Metaverse the next internet, with the media dutifully covering it as if it were real along with consultancies like McKinsey and Deloitte. 

To be clear, there was never proof that the Metaverse was a real thing that anyone wanted, but because Facebook changed its name to Meta, the assumption was that the rich and powerful would not simply “do something for no reason.” It petered out because despite all the hype, there was very little to actually use, or invest in, or really do with it.

Yet the most important detail is how everybody went on with their lives and ignored that Meta burned $77 billion on nothing, [Zitron actually under states Meta's losses here which continued to accumulate after the $77 billion figure became public. -- MP] or that Microsoft (which bought Activision Blizzard under the auspices of the Metaverse) CEO Satya Nadella said he “could not overstate the breakthrough of the metaverse” in 2021 and then effectively shut it all down by 2023. Nobody was fired for the fuckup. Nobody got in trouble. No media outlet apologized for being wrong. As regular people were fired thousands at a time, tech executives like Satya Nadella and Sundar Pichai received tens of millions of dollars a year. Everybody acted like nothing happened.

Put another way, when regular people fuck up, they see themselves restricted and punished, fired, their credit ratings dropped, evicted from their houses, embarrassed in front of their friends and peers, and when corporations fuck up, the system accelerates to isolate them from damage, explain away their faults, congratulate them for trying, and then forcefully return “reality” to a point where everything they say is perfect. 

 Given that these very same people, boosted by the same publications, are spending not hundreds of billions but trillions of dollars, putting the markets and possibly even the economy at risk while doing considerable damage to the environment in pursuit of their next next big thing, it seems like their last couple of fiascos should come up in conversation a bit more often.

 

It's like something from a science fiction story. (No, it's literally* something from a science fiction novel)

[* Can you use literally to describe something being derived from something even if the result is not a direct quote? Well, I just did so apparently the answer is yes.]

This makes perfect sense when you think about it. Large language models are designed to generate plausible strings of text with respect to their training data. Think about all the times you've heard chatbots like Claude or ChatGPT compared to science fiction or described in sci-fi terms. If you ask an LLM what it's "thinking," it is not at all surprising that it would reference the sci-fi books and plot descriptions in its training data when coming up with a reasonable-sounding answer, particularly when you take into account the noise that comes with the feedback loops of agentic systems. 

From Cal Newport: [emphasis added]

OpenAI was looking at these reasoning traces to find unnerving examples of plotting behavior from its “swarm.” This seems like a natural thing to do, but there are two problems with this approach:

1. These chain-of-thought traces don’t necessarily reflect the actual logic behind an LLM’s ultimate answer or suggestion. Multiple studies have shown that these models sometimes invent reasoning that sounds plausible, but may be completely unrelated to how they arrived at the response. (See, for example, ​this paper​ from ICML 2026, or ​this paper​ from NeurIPS 2023.)
   
2. Research has also shown that referencing the fact that an LLM is an AI system in a prompt increases the chances that the LLM’s output will reflect sci-fi style narratives about AI running amok. Because it was trained on many such stories, the model assumes that this is the type of output it’s supposed to produce. If you take sci-fi tales out of a model’s training set, it’s less likely to talk in terms of AI running amok. (See, for example, ​this study​.)

Put these two observations together, and it’s clear that it borders on research malpractice to soberly report on carefully curated clips from these traces to imply that somehow the combination of these prompt loops and the LLM they are prompting is a unified sentient entity with malicious intent. It’s more likely that the LLM in question is simply post-hoc rationalizing its outputs with well-worn tropes it encountered during training.

Though much more difficult to study experimentally, I wonder if this might explain how frequently sci-fi and fantasy elements show up in cases of AI psychosis. 

Coming soon to a theater near you...

I'm sure Musk will deal with this with the grace and dignity we've come to expect.  

Steven Zeitchik writing for the Hollywood Reporter:

Over the course of a decorated career, Gibney, 72, had taken on Scientology and Enron, the U.S. military and Vladimir Putin, the Catholic Church and Wall Street. All of that prepared him, though nothing could really prepare him, for this magnum Musk creation — cinema’s premier curtain-puller on power revealing the man who has amassed so much of it. When it comes out this fall, Musk will look to change the game in several crucial ways, lining up a target that Hollywood never shoots at, seeking to become a hit documentary in an era allergic to them, and trying to open the eyes of American voters to a key Republican electoral arms dealer who Gibney says is mounting an unfair fight. At stake with Musk, its creator wants to tell us, is nothing less than the future of democracy. And to spread that message, he has enlisted some of the biggest media companies around — including, improbably enough, the one soon to be owned by Trump ally David Ellison.

...

So much Musk have we ambiently ingested these past few years that perhaps the best cultural gift might be a break from all that, an Elon-free zone populated with anyone besides the 55-year-old provocateur.

Yet Gibney aggressively wants to challenge that instinct, and he may have a point. None of what we have seen and thought about Musk can fully ready us for what the filmmaker has done here — an adventurous but methodical movie grounded in the thesis that, as he sees it, a dangerous huckster is stealthily aggregating power to exploit us, primarily by centralizing and controlling massive amounts of data. Gibney’s basic take on Musk is that he’s a confidence man whose power comes only from the stock market and disappears the minute we see through the facade of sales pitches — but can do a whole lot of damage until that happens.

The film world-premieres Sept. 8 at Venice. But its cultural reach will likely extend far beyond that. Musk stands poised to touch off a fresh wave of anger and fear among the millions already inclined to worry about him, into and well after the midterms, while fueling a whole lot of trollishness (and worse?) from the many who worship at the temple of Elon Musk, including Elon Musk.

Gibney has crafted a movie that, while not new in its specifics, is often damning in its totality, standing apart from anything the media entertainment complex has recently turned out — a portrait of power with the cinematic scope of All the King’s Men and the potential social impact of The Pentagon Papers. (“It’s rare these days to see a complete story instead of the fragments we’re used to,” says Anonymous Content’s Nick Shumaker, a producer on the film.) Some will surely find Musk overly negative and conspiratorial. Yet nearly every time the film seems like it might tip into paranoid fantasy, Gibney unearths a new fact or asks a new commonsense question to steady us back.

 

Too much to summarize but I do have to share this one anecdote:  

The insinuations of business hucksterism are one thing that could provoke him, more intimate details are another. Musk’s personal life comes in for tough scrutiny in the film, ... with a revealing childhood story Musk’s father tells in which a family friend undergoes a medical emergency while they were all on a road trip in Musk’s native South Africa. The man dies as everyone tries to resuscitate him. Young Elon, his father recounts, reads a comic book, unreactive through the entire episode. 

 

It's Labor Day, so we're taking time off and running a repost

 

It's Labor Day weekend so let's stop by Emma Goldman's ice-cream shoppe in Worcester for a hot fudge sundae and some anarchy. pic.twitter.com/LgPLuxkmWG

— Charles P. Pierce (@CharlesPPierce) September 4, 2021

Look for the Union Label

The ILGWU sponsored a contest among its members in the 1970s for an advertising jingle to advocate buying ILGWU-made garments. The winner was Look for the union label.[9][10] The Union's "Look for the Union Label" song went as follows:

    Look for the union label
    When you are buying a coat, dress, or blouse,
    Remember somewhere our union's sewing,
    Our wages going to feed the kids and run the house,
    We work hard, but who's complaining?
    Thanks to the ILG, we're paying our way,
    So always look for the union label,
    It says we're able to make it in the USA!

The commercial featuring the famous song was parodied on a late-1970s episode of Saturday Night Live in a fake commercial for The Dope Growers Union and on the March 19, 1977, episode (#10.22) of The Carol Burnett Show. It was also parodied in the South Park episode "Freak Strike" (2002).

"An analyst at JP Morgan described the approach as being like paying your mortgage with your credit card, which is unkind, but it's also basically what it is."

I realize we recommend a lot of Patrick Boyle videos but this is one of the best walk-throughs of one of the most important news stories of the moment (and 2026 has not been notable for slow news days).

From the video description:

Treasury Secretary Scott Bessent is trying to force down US Treasury yields with surprise bond buybacks — and it isn't working. In this video we break down Bessent's activist debt management strategy, why doubling the Treasury's long-dated buybacks is a bet on falling interest rates funded by short-term bills, and why his old boss Stanley Druckenmiller publicly tore the plan apart in a Wall Street Journal op-ed ("Let the Bond Market Speak"). We look at the collision with new Federal Reserve Chair Kevin Warsh after Jackson Hole, the 50% tariffs on Canada and the Mark Carney feud, "Operation Economic Outcast" and the secondary-sanctions problem with China and Iranian oil, the GENIUS Act and crypto's role in sanctions evasion, and Stephen Miran's case for the defense. The through-line: you can't trade around arithmetic. When a government goes to war with market prices, the bond market has an infinite balance sheet — and prices tend to win.

Elites in America

This is Joseph. 

This article got me thinking about the idea of elites. Because I am actually in agreement with one major thread here, which is that the focus in American life has been on intra-state competition and not inter-state competition. This shift in focus to internal competition has often gone poorly for states. Stalin weakened Russia versus the Nazi's by seeking internal purity. Byzantium lost to the Ottomans because they couldn't overcome internal strife. It is generally not a good state of affairs.

But I worry that we've move to an era where this good sense has been lost. For example, what national interest has the trade war with Canada served? Do any of the proponents actually think that it improves matters to put crushing tariffs on a neighboring country? Has anybody held this position prior to 2017 and what would have changed about the fundamentals in 2017? 

What is very much true is that everybody can identify elites that are acting poorly, at least from their point of view. I am even tempted to argue that some of them really are, mostly because of spillover from the culture, probably reached this status objectively. 

But what is really missing is focus on trying to improve obvious things in America. Now, you probably think the next words out of my mouth will be some variation on "both sides should", but I do think it is an actually asymmetrical problem. One one side you have people who have been . . . tentative about holding strong opinions or fighting. The other side appears to have started a war for no clear objectives. Or, if there are objectives, they are amazingly well concealed from everybody who has been thinking about the war. 

And Iran is a good place to end. It's cost a ton of money just as the social security fund is going to need some legislative tender loving care. It has lead to bases being bombed, Americans being killed, and there really isn't a great view of an exit strategy that doesn't start undermining the world's trade infrastructure. Bret Devereaux has a great piece on this topic (and I fear I might be part of the "Eye of Sauron" he was worried would be drawn to this piece) and it's pretty clear that everybody is worse off if the modern trading state is simply broken. 

It's also possible for the collapse of an unjust system to leave everyone worse off. Ask the Italians about the fall of the (slave holding) Roman Empire how great the aftermath was. 

So I think it is important that we go into the fall midterms with some focus about what the stakes are and how we can pull things back towards a better trajectory. 

Will Fusion Fizzle?

Following up on our previous post.
 
As is always the case with technologies that are still on the horizon, reasonable estimates of timelines and economics vary widely with fusion power. Unfortunately, the dominant voices in the discussion—the people journalists are most eager to cite and treat as authoritative—are some of the least trustworthy people you'll find, operating under some of the largest conflicts of interest imaginable. Pitchmen vying for billions of dollars in venture capital are treated as reliable and independent experts on the subject, telling their stories to journalists who never saw a gee-whiz, the-future-is-now story that they didn't rush to print. Take Hellion which basically pulled an operational date of 2028 out of its ass. In case you're curious, orifice-derived estimates will get you around $1.5 billion in today's market. 

On top of that, we need to remember that the push for fusion startups is tied in with a highly representative example of the techno-optimist narrative, right down to the shaky economics and the tendency to substitute a fascination with postwar pulp sci-fi for actual engineering and economic analysis.

In terms of the engineering, if you look at experts not on the payroll of these companies and independent of the larger AI/robotics bubble, you'll find few who predict viable fusion reactors five years from now and many who think we are looking at 20 or 30 years in the future.

In terms of the economics, it is far from certain that fusion power will be cheaper than fission—remember, the majority of the cost of nuclear power today is on the operating side, not the fuel—and based on most estimates, it is likely to be more expensive than renewables such as solar and wind. This isn't to say the technology isn't promising and worth pursuing, but the stories we are being told about it are largely fantastic, and the surge of VC capital flowing into the sector makes no sense.

There is one more factor to take into account. The most powerful men in tech—Musk, Altman, Zuckerberg, et al.—are currently sitting on top of a multitrillion-dollar AI bubble that requires ever-increasing data center construction. According to some analysts, we are building far more capacity than we need, but if growth levels off, it threatens to deflate the entire bouncy house. One of the consequences of this unprecedented infrastructure expenditure has been a tremendous strain on our power grid and a huge surge in greenhouse gas emissions. Promising that a still-nonexistent technology will emerge in the next five or six years to solve the problem is extremely convenient for these people. It is also right on brand.

From "Notes on the Recent Hype about Imminence of Commercial Fusion Energy"

by John P. Holdren (who is kind of a big deal in the field) of the Belfer Center for Science and International Affairs

Originally published in June 2024 and updated in January 2026.

The best performances in fusion experiments to date—both in a tokamak and in the NIF―have been achieved using the deuterium-tritium (D-T) reaction. Getting a true energy gain from this reaction—more fusion energy out than the energy input to the system―is inherently much easier than achieving that with any other known fusion reaction; but it is nonetheless so difficult that nearly seventy years of research on controlled fusion worldwide, at a cost of many tens of billions of dollars, have not yet led to reaching this threshold of so-called “scientific feasibility.”

As I will explain, working with tritium poses big challenges for reactor design, for maintenance, for worker safety, and for public radiation exposures, but the difficulty of achieving even bare energy breakeven with any fusion reaction other than D-T is so great that most fusion scientists believe that success with reactions that avoid tritium will come much later if it comes at all. In what follows, I first address the challenges in making a practical magnetic-confinement fusion reactor using the D-T reaction, then the even bigger challenges of doing so with a laser-driven system using D-T.

...

Bombardment by the 14-MeV neutrons produced by D-T fusion weakens most structural materials in short order, as well as turning steel, molybdenum, titanium, and most other metals intensely radioactive. The first problem could require frequent replacement of structural components exposed to fusion neutrons (quite possibly so frequent as to make the whole operation uneconomic) and the second problem greatly complicates all maintenance around the reactor innards (further threatening economic viability). How many of the today’s fusion optimists have thought about the stresses that operating continuously will impose on equipment working at the boundaries of conditions that today can be tolerated only briefly…or about how this challenge could be surmounted?

Tritium—the radioactive isotope of hydrogen—is almost nonexistent in nature. It is produced for today’s uses by neutron bombardment of lithium inserts in fission reactors. The large quantities needed for D-T fusion-reactor operation would need to be “bred” in the reactor by bombardment of lithium in the reactor blanket by the 14-MeV neutrons. Tritium’s radiation is less penetrating than that of the main activation products of concern, but as an isotope of hydrogen it is comparably volatile, and the quantities stored and flowing in and around a D-T fusion reactor would represent a significant additional radiological hazard. The degree of tritium control needed to stay within current guidelines for public radiation exposure at a nuclear plant boundary would be extremely challenging to achieve in a tritium system as complex as that in a fusion reactor.

The significant environmental/safety issues posed by tritium and activation products (which, by the way, will likely require long-term management as radioactive waste) have been ignored by practically all of the recent commentators, most of whom insist in passing that fusion is "clean."

...

But the factors by which the repetition rate and pellet cost in the NIF fall short of what would be required in a reactor are far larger. Today, with a large crew of highly trained specialists in full-time attendance, the NIF is lucky if it can get off two shots per day. A practical laser fusion reactor would require something like ten shots per second. As for the cost of the pellets that the laser beams irradiate, when last I looked they cost in the range of $10,000 each. For a practical fusion reactor, that cost would need to be no more than a few cents apiece. 

The enormous challenges posed by the use of tritium in magnetic-fusion reactors would not be less, and might be even larger, for inertial-confinement reactors. 

I think it’s fair to say that the scientific and technological problems that must be solved on the road to a practical fusion reactor—beyond energy breakeven in sustained operation—are not much easier than the breakeven problem itself, which is still unsolved after 70 years of costly international effort.

Furthermore, there is no guarantee that, if all the technical challenges for either the magnetic or the inertial-confinement approach could be surmounted, the result would be economically competitive with other relatively clean energy options available in the same timeframe. The frequently heard claims about fusion being “cheap” take no account of the cost of building the reactor itself. The raw fuel for fusion would be cheap, as it is for fission reactors today. But the cost of the energy from fission is dominated by the cost of building and maintaining the reactor and associated facilities, not by fuel cost, and the same will be true of fusion. The studies of the likely economics of hypothetical fusion reactors in which I’ve been involved all indicated that it will be a great challenge to make electricity from fusion less expensive that that from fission.

...

Nobody has a clear crystal ball when it comes to the characteristics of future technologies, and I am further handicapped by not having followed fusion science and technology particularly closely since leaving the White House in January 2017. Still, if I were a betting person and if I expected to be alive to collect, I’d bet we won’t see a successful commercial fusion reactor before 2050.  

 

Fusion hype has a familiar ring to it

We'll be coming back to this Holdren essay and what it says about techno-optimists' fascination with fusion, but for now, but for now, there's one aspect of one paragraph I'd like to focus on. The hype/investment bubble around companies like Commonwealth and Helion is very much a part of the tech story of the 2020s, including the most troubling aspects. If you replace fusion in this paragraph with AI or with humanoid robots, you will still have a statement that is both accurate and timely.

From John P. Holdren of the  Belfer Center for Science and International Affairs
(Originally published in June 2024 and updated in January 2026.)

But the recent hype about the chances for early success in the Earth-bound electricity-generation role has been far over the top, in my view, and is dangerous in the sense that large early investments in the false expectation of early commercial returns—which would be only the latest in the long cycle of overoptimism about fusion followed by disappointment―could well set back the timetable for eventual success, not least by locking in technological dead ends in premature pursuit of commercialization. The hype is dangerous, as well, in feeding the false hope that there is a technological silver bullet that will save us from the challenging problems of addressing climate change with the means now available or easily foreseeable. 

 

With all three technologies, tremendous investments have been made based on assumptions of imminent massive ROI, assumptions that range from questionable to highly unlikely. In all three, this combination of capital expenditure and hype has locked research into questionable paths (see our previous post on steam-powered airplanes). With artificial intelligence, large language models, a promising and impressive technology but one with notable limitations, have sucked much of the oxygen out of the room when it comes to pursuing other types of AI. With humanoids, the situation is far worse. This is a branch of robotics that was relegated to little more than a curiosity a few years ago and which owes its success to an embarrassing publicity stunt where Elon Musk brought out a dancer in a robot suit. 

We looking at the largest technology-driven market bubble in human history, but we are also seeing potentially disastrous public policy being justified by promises that the tech is unlikely to ever fulfill. The chairman of the Fed assures us that the coming wave of productivity will fix our economic woes. Op-ed pieces tell us that AI can help solve the problem of climate change. Silver bullets for everybody any day now.