OpenAI

On Wednesday, the AISI, which evaluates AI models for the British government, said both Anthropic’s Claude Mythos Preview and OpenAI’s GPT-5.5 showed progress well above previous trends on cybersecurity testing. Separately, XBOW released data suggesting “frontier models have taken a major step forward in vulnerability discovery.”

U.S. District Judge Yvonne Gonzalez Rogers, the judge presiding over Musk v. Altman, had told Musk when he left the stand that he was not excused from the trial and that he was still under “recall status,” meaning he should stay nearby and ready to testify. But he’s currently in Beijing…

There is no rebuttal case from Musk’s team. We will get closing statements tomorrow.

It was on “functional expsenses,” ie, salaries, compute, etc. The cross is just arguing about the methodology of accounting for commingled money in the donated accounts. I can’t believe we are having a methodology dispute about this. I may die.

John Coates noted that he’s worked for a lot of law firms as an expert witness, including Quinn Emanuel, Musk’s primary firm — and not the one trying the case today. He is excused. The judge is now huddling in sidebar with the primary lawyers for the case, and an animated discussion is taking place.

Also, he apparently has worked as an expert witness on a few Twitter cases, including the one where Musk tried to get out of buying Twitter. Incidentally, OpenAI’s lawyers are also the ones who made Musk buy Twitter. Is that deliberate shade? Who can say.

Some highlights:

So during the opening statements, Musk’s lawyers said that a for-profit like a museum gift shop shouldn’t be bigger than a nonprofit, like a museum. We are now hearing from Daniel Hemel, OpenAI’s expert witness. Guess what? Museum gift shops generally aren’t for-profit; they’re part of the nonprofit. Also, OpenAI’s for-profit isn’t ancillary to the nonprofit — it’s how the nonprofit pursues its mission, like with the Mozilla Foundation and the Mozilla Corportation.

He said that “for a large nonprofit organization, having for-profit affiliates is very much the norm.” When asked, he also said that oftentimes, the for-profit affiliate of a nonprofit is “quite large compared to the nonprofit,” and he gave the Mozilla Corporation (which owns the Firefox web browser) and the Mozilla Foundation as an example. Hemel also testified that he’s getting paid $1,750 an hour to be here.

He said Musk was concerned about Google DeepMind and CEO Demis Hassabis and “expressed a lot of concerns about what would happen if DeepMind got to AGI first.” Achiam said he shared his concern that trying to “race” towards the technology was a “fairly unsafe proposition … He was proposing to do something that seemed … obviously unsafe and reckless.”

She quotes a tweet of his saying that he believes Musk was doing his best for humanity. He asks when that was. She says, January 2025. He says, well he’s done some things that undermined my confidence since then.
There’s a brief redirect, and then Achiam steps down. No trophy for the jury. :(

“Are you aware that OpenAI employees are better-compensated than any other employees in startup history?” lol lady, why would he know that. Anyway, he’s got millions of dollars in OpenAI shares, and he’s also sold some for more than $10 million.

In Musk’s testimony, he claimed he might have said something friendly like “don’t be a jackass” but denied he’d called anyone a jackass. Achiam’s testimony obviously contradicts that. Achiam received a trophy from Dario Amodei at the next meeting in commemoration of Achiam standing up to Musk: “Never stop being a jackass for safety.” The trophy is not introduced, sadly for me.

“It sounded like he wanted to race toward AGI.” That sounded unsafe to Achiam. “He was proposing to do something that seemed, based on our understanding at the time, obviously unsafe and reckless,” Achiam said. “We had a pretty tense exchange, and he snapped and called me a jackass.” There were 50 or 60 people at that meeting.

He had a notable interaction with Musk, though, during the all-hands when Musk was departing the organization in Feb. 2018. Musk explained that he was leaving because he had a new conflict of interest with Tesla, which would be hiring from the same pool of researchers — and indicated a general lack of confidence in OpenAI’s path

That’s according to Josh Achiam, currently the company’s chief futurist, who joined in 2017. He said Sutskever’s impassioned speeches would typically be about the science-fiction-esque future that was approaching.

He said Brockman and Sutskever were the “main leaders,” and that Brockman was the “engineering workhorse that pushed to build scaled-up systems that would train the AI and make it work.” Achiam called Sutskever a “scientific visionary” who articulated what the future would be like, such as football fields of silicon chips making large-scale calculations.

He said when he joined, OpenAI was a team of about 50 people, and that it essentially felt like “an extension of a graduate student lab in a university” — a “collegiate, academic, super intellectual” environment — with most employees being either current PhD students or recent graduates. He said he appreciated that there wasn’t a “publish or perish” type of culture at the time.

His job was safety research then. He is now the “chief futurist” at OpenAI, where he tries to think about side-effects of AI (such as social impacts, economic impacts, and consequences for national and international security). “It is my best attempt to have us fulfill the mission of OpenAI,” he says. The idea is to ensure AGI benefits everyone, he says. It’s “one of the highest and noblest callings we could possibly have.”

He is establishing his background right now. You will be just shocked to hear that he’s into science fiction. This is the witness we may see the jackass trophy for. I am on the edge of my seat.

Microsoft had an approval right on some transactions. It did not have the majority of the board. That’s even though they contributed more than 90 percent of OpenAI’s initial investments. Also, all LPs had major decision rights, Wetter testifies. So this is less control than Musk wanted for more money.

“We did not talk to Elon Musk during out due diligence process,” Wetter notes. He’s not a party to OpenAI’s agreements with Microsoft. A lot of the direct was “Are there any agreements with Elon Musk here? Are there any there?”

We have just gone through the terms of a very boring document. I will spare you. That’s the top line.

He lead corporate development at Microsoft, where he’s worked for almost 20 years. We saw this deposition earlier as part of Musk’s case. He did a bunch of the work on the 2021 and 2023 OpenAI deals. I believe he is here to talk about Microsoft’s due diligence and also to put the deal in context — “we’ve done over 100 transactions including acquisitions and investments,” in aggregate value of $100 billion.

He also doesn’t remember a bunch of things Musk’s lawyer is asking about. I fully believe him on this — feels like Scott’s only real interest is the tech. He was so happy talking about Azure and he is very lost talking about partnership agreements.

She seems confused by a CTO not knowing what revenue had been generated. Scott noted he was not the chief revenue officer. He seemed amused.

He has testified that the company liked the idea of partnering with OpenAI in part because it would show how to build out Azure for AI frontier research. It’s pleasantly boring.

He said he sometimes used strong language at work, but might have said something like, “Don’t be a jackass.” So in addition to being hilarious, the trophy also makes him look like a liar.

There is a trophy that OpenAI has brought in, that’s half of a donkey — the back half — and says, “Never stop being a jackass.” It’s a commemoration OpenAI employees bought for another employee that Musk called a jackass on the way out on his last day. Musk’s team does not want the trophy in evidence.

It joins a handful of other tech companies like Snap and Microsoft in supporting the bill, while major tech groups maintain opposition. The announcement comes as a key Senate committee prepares to move forward on its version of KOSA, after a House committee passed a largely overhauled version.

Kolter laid out OpenAI’s different safety groups: the safety systems team, which works on guardrails and evaluations; the preparedness team, which deals with OpenAI’s preparedness framework; the alignment team, which helps train models on ways that “align with human values”; the model policy team, which develops the model spec; and other teams focusing on investigations. When speaking about the controversial dissolution of OpenAI’s superalignment team and AGI readiness team, he said some of that research is being done by other teams.

Dr. Jeremy “Zico” Kolter said that so far, there have been two times when the committee “formally requested a delay of models.”

Anyway, he’s a member of the nonprofit board of the OpenAI foundation, but not the for-profit. He’s a safety expert.

He’s a Carnegie Mellon prof who focuses on safety and security.

Altman said — and evidence in the direct showed — that Altman had made sure Musk knew about Microsoft. “I would often have to remind Musk of things, but this one I assumed we had talked about enough times that he would remember,” Altman says. Of Musk’s attitude toward OpenAI, he says, “we were kind of left for dead.”

And that Musk hasn’t sued any of them. Softbank’s investment is bigger — so is Nvidia’s and so is Amazon’s. That’s all from Microsoft.

In response, Altman texted, “I agree this feels bad — we offered you equity when we established the cap profit, which you didn’t want at the time but we are still very happy to do any time you like.” Molo tries to ask if this is a bribe — if Altman is trying to say that if Musk lets him get away with stealing from a charity, he’ll split the loot. Savitt objects, YGR sustains.