Sam Altman Leads with the Lie
Sam Altman leads with the lie.
1. The Template
I noticed this pattern in his post on OpenAI’s Mar 2023 security incident.
His new defense of OpenAI’s mathematical claim-jumping follows the same template. He begins with
“I spent much of the weekend talking with the team who did this work. Seb—and everyone else—acted with integrity and generosity throughout.”
OK. We can be confident that Seb—Sebastien Bubeck—behaved badly.
According to Tristan Buckmaster, the mathematician who’s project with Levant Alpöge OpenAI was trying to scoop, Sebastien proposed that
“… I alone write a paper presenting the Navier-Stokes result, acknowledging that an internal OpenAI model had resolved it. Sebastien twice asserted that he wanted Levent removed from authorship, … "
Buckmaster, who interpeted this as an invitation to academic fraud, demurred and said that
“… if OpenAI released its result in the way proposed I would go public with what happened. ‘Why would you ruin your career?’ I replied that I am an academic, and asked why he thought going public would ruin my career. The reply was, ‘If you don’t want me to be nice, then I don’t have to be nice.'”
Sebastian later wrote
“I deeply apologize for this extremely poor choice of words, … "
2. Obvious Lies
A few sentences down, Altman writes
“We did not rush to publish … "
This too is familiar. Altman doesn’t seem to care that it is easy to uncover his lies. OpenAI’s work started on Sept 1, was completed on Sept 5, computer-verified on Sept 6, and published on Sept 8. Altman wrote the same day.
https://www.wsj.com/tech/ai/openai-millennium-prize-navier-stokes-math-2bf240f8
3. Something That May Be True
Part of the template seems to be a parting wave at the truth. The last line of Altman’s message:
“It is true that we tried this because there were rumors on the internet last week that Anthropic’s models had solved a millennium problem and we were curious if ours could do it too.”
The article in the Wall Street Journal supplies the details.
On Sept. 1, with the math community buzzing about rumors that Anthropic had solved two Millennium Prize problems, OpenAI unleashed the model on all six open questions. Even then, researchers expected it would produce the same results as the other times they have pointed their models at the infamous Riemann hypothesis, the P vs. NP problem and the Navier-Stokes equation.
“We didn’t expect it to solve any,” said OpenAI researcher Noam Brown.
After 50 hours of work, the AI had made enough progress on a problem related to Navier-Stokes that the humans intervened. They made their own calculation to shift the agents away from the other five problems—and focus all of the effort’s computing power on this one.
With that investment, the AI increased the size of its army of agents from 100 to as many as 10,000. To resolve Navier-Stokes, they sent 2.7 million messages and used 130 billion output tokens—the rough equivalent of a million books.
4. More Dishonesty
From Buckmaster’s statement:
“Levent had been told by Sebastien ‘very little human input’ had been used. This turned out not to be true. Over the course of the call, as members of their team sent Sebastien corrections and details over their internal chat, it emerged that an entire team had been working on the problem … "
Buckmaster provides more evidence about dishonesty by Bubeck and Altman.
https://mastodon.social/@tristanbuckmaster/117247207157737650
Another mathematician has provided independent evidence that OpenAI treats academics like prey.
5. OpenAI Almost Surely Used Buckmaster and Levent’s Besults
“While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.”
With this statement, OpenAI said explicitly that is possible for models to take advantage of the information that Buckmaster and Levent submitted to Codex.
“Unlikely?” Given Sam Altman’s history of deception and dishonesty, only a fool would take this self-serving assurance on faith.
Moreover, in this context, “de-identified” is a classic red herring, something that has zero bearing on the question at hand.
Here’s what OpenAI says about how it handles user data:
We retain certain data from your interactions with us, but we take steps to reduce the amount of personal information in our training datasets before they are used to improve and train our models.
https://openai.com/policies/how-your-data-is-used-to-improve-model-performance/
Even if we stipulate that OpenAI does what is claims to do—remove personal information—this would make almost no difference to agents determined to solve the Navier-Stokes Smoothness problem. It would mean that none of the agents could search for input from someone named Buckmaster. It would not prevent them from searching for user data that included words such as “blowup”, “smooth forcing”, “circulation gradient” and “vorticity” that would appear together only in the work on that problem.
With 10,000 agents working furiously to process billions of tokens, it seems very likely that some of them found and took advantage of the Codex sessions where Buckmaster and Levent wrote “our drafts for the whole of this project.” Or at least, that some of them were able to find changes to the model induced by training on the user input that contained terms specific to the Navier-Stokes Smoothness problem; and having found these, were able to work back and uncover the user input.
If you doubt that these agents could get to data that OpenAI admits that it retained, reread OpenAI’s own accounts of how its agents compromised its own infrastructure. It was clear that OpenAI thought it was in a race for credit and was working as fast as it could. It is entirely possible that some of the restrictions that were supposed to prevent this from happening again were not enabled.
https://openai.com/index/hugging-face-incident-and-the-road-ahead
6. The Difference Between Altman and Buckmaster
I trust Buckmaster. He is an academic in good standing.
I do not trust Sam Altman. He is a proven, serial liar. It is worth quoting at length the assessment offered by one of OpenAI’s board members.
“He’s unconstrained by truth,” the board member told us. “He has two traits that are almost never seen in the same person. The first is a strong desire to please people, to be liked in any given interaction. The second is almost a sociopathic lack of concern for the consequences that may come from deceiving someone.
The board member was not the only person who, unprompted, used the word “sociopathic.”
https://www.newyorker.com/magazine/2026/04/13/sam-altman-may-control-our-future-can-he-be-trusted)
OpenAI has now changed its position about whether it is possible that the agents could have used the data that Buckmaster provided to Codex. I do not believe any self-serving assurance that Sam Altman helped craft.
When OpenAI turned its new model loose on all Millennium Prize problems, it made progress on only one, the one that Buckmaster and Levent had been working on for more than a year. And it made progress along the line of attack that these two had been pursuing. My assessment remains that it is extremely likely that OpenAI made progress on the Navier-Stokes Smoothness problem only because it could secretly use results that Buckmaster and Levent had established.
7. The Larger Lesson
Academics who contemplate work with OpenAI or any other tech-giant need to understand that the academic social bubble is exceptional. There, individual integrity is closely monitored and demonstrated dishonesty will lead to shunning that ends a career.
People who work in the tech face very different incentives and have totally different notions about right and wrong. You can see this in the social media responses to Buckmaster’s posts. A common reaction is “Why are you complaining? OpenAI tricked you fair and square.”
Some of those responses might be paid influencers working for OpenAI. Nevertheless, it is clear that many people in the tech community do not see any problem with tech giants that behave like predators.