“The document says the signatories believe each company should implement controls and audits, believe, should. That's not a commitment. That's a mission statement. … No CIO signs a seven-figure deal because a vendor posted a pledge. They want evidence. Safety should clear the same bar.”
The Six Five with Patrick Moorhead and Daniel Newman · October 2026
This came in the show's debate segment, where the two hosts take opposite sides. The topic was a voluntary AI safety accord signed by the large AI companies, and Newman had the side against it. His co-host had just argued it was real progress.
Transcript
billion. She's finally getting paid, right? Isn't she the AI Writes of OG that hand labeled the first frontier or first former model and gave it all away for free, made no money. So she's getting paid now. Her and her partners. I mean, look, I guess you could call it the two most powerful women in AI have now teamed up to work together. And, you know, on the AMD side, like, look, this is what a trillion-dollar balance sheet enables you to do is make $8 billion bets. And that's what this is here. Like, this is really a bet on kind of what's next, world models, quantum. These are the kind of areas with physical AI of like where are the next trillions of dollars of TAM sit. And that's, you know, not all AI is in the data center, that labeling the physical world, for instance, is going to be a challenge and a task that's going to need to be addressed in the long term and will be driving the future of the frontier. And so to me, this feels like a big buy-in to physical AI, but it also just feels like a big talent grab, in my opinion. It's a massive investment, but to grab what would be seen as probably the foremost talent in this particular space. And, you know, I think, you know, the economics are a little bit less well known. Like how much actual economic value is there to this deal? But again, you know, Lisa can play with the stock. I mean, it's, it's like I said, the company's a trillion dollar value now. This is like Elon doing the cursor deal. It's just like, I can do this. I'm, you know, I'm just going to ramrod this right into this massive valuation of the company. I'm going to buy interesting assets and parts and pieces and talent and people, and I'm going to put it together because we need to be building for what's now and we need to be building for what's next. But, you know, it's a very interesting move. Like I said, it's just, there's almost no numbers around it. Like there's no numbers around what they own, what the business is, what the revenues look like. This is a speculative play into a future ham that I'm sure AMD feels it needs to be competing in.
Yeah. So AMD was an investor in this company. Feife was on stage at AMD's AI event as well. So the two know each other well. By the way, NVIDIA was an investor as well. So a couple of people. Takeaways. I think the most that I got, I got a QA with Vomsi Bopana, who runs AI at AMD. And here's kind of my net on this. This is a talent and model insight buy. It's not a revenue play. You called it zero revenue. Maybe they got some NRE from some of their partners, but likely zero revenue. It's funny, it also got priced at 1.6 times the $5 billion valuation that Bloomberg reported. And the other thing, and I know people jumped on this, and I would understand if you read Fei Fei's blog, it's like this is a physical AI play. But in my QA with Vomsey, he was very clear. This scope is AMD-wide. It's horizontal model capabilities for the company across content, media, robotics, and embedded in data center prediction and reasoning. So I think that's, you know, I appreciated his clarification here. The key here is, and I also brought this up in my call with him, is this team has to build models to keep that talent there and to keep that talent and to attract talent as well. So it's interesting. I don't know if AMD is going to release competitive models and where they're going to release it, but I'm excited to say what I can say without hesitation is this team, if they can keep the core team in there, will absolutely give better insights into what the hardware roadmap should look like as well. All right, folks, that is the decode. Why don't we jump into the flip now? So we talked a little bit in the decode about the White House AI Accord, but the question is, will it solve real problems? Or, you know, or was this just a great photo opportunity for the United SACE of America? Let's jump in. All right. I am for this. This is going to make practical progress before we get into new legislation. So, first of all, getting competitors behind all the response, common responsibilities can make it really hard to dismiss safety as someone else's problem. I think that's a good thing, right? Independent evaluation, board-level review offer concrete mechanisms and not just merely a statement that AI should be safe. And it's certainly better than coming in and saying, regulate me or recuse me from having any product liability if my agents come and end up killing you or something like that. So the self-governance is where this is heading, right? And it's what the labs actually wanted, aside from getting that liability, get out of jail card-free. And I do like, you know, when you combine this with stuff that NVIDIA brings out that actually has an engineering solution over the problem, I feel pretty good about that. So net net, I just think that this wasn't just a photo op and there were those real teeth in here. And quite frankly, I think the behavior is already changing. I mean, I know we were kind of joking in the decode about these new models coming out. The two models from the frontier labs in question, Anthropic and OpenAI, was really about cost reductions. These were smaller models to give more capabilities, though they weren't actually pacing anything on the frontier. We already saw OpenAI pause frontier training after its containment failures. And by the way, a final comment here, there is a backstop here, right? The FTC was investigating before the accord and confirmed the next day. So the existing authority still applies. And a signed public standard, I think, gives enforcers and buyers something. Specific to hold to hold companies to.
Yeah, Pat, I'd love to agree with you, but then we'd both be wrong. You know, Mark Zuckerberg, the face of this new pledge and finding out their very cool new Muse agent is hacking people's eye messaging, taking something that's supposed to be extremely secure and giving away people's addresses that are using Muse to transact in different platforms. It's very safe. Get people access to my home and all my most private interactions with the people closest to me. Very, very good stuff. But in serious, like, you know where I stand. I said it this week. If you build the product, you're responsible to make it safe and secure. And that's why the accord just doesn't do it with me. Do it for me. You know, start with the language. The document says the signatories believe each company should implement controls and audits, believe, should. That's not a commitment. That's a mission statement. Second, there are no teeth. Speaker called it voluntary. The president called it morally binding. Congress didn't announce any way to enforce it. And the accord itself admits none of this is law. And what's the oversight that was floated? A committee of maybe 10 people that could come from the same group that signed it. You know, that's basically players refereeing their own game. Third, what's actually new? Open AI and Anthropic had already committed to third-party evaluators before anyone sat down for lunch. So tell me what changes. Monday morning, which model launch gets held, which agent loses permission, which board votes no. No one can answer that from this. So the final part is like, look at the incentives. Same day this was signed, the president said he'd never stifle the growth of this technology. So the message to Mark is pretty clear. The pledge is the guardrail. And, you know, look, we're in a hard ROI era. No CIO signs a seven-figure deal because a vendor posted a pledge. They want evidence. Safety should clear the same bar. Companies playing offense on safety, ship evidence. Companies playing defense, ship photo ops. I'll give Zuck credit. He called it a start, but a start isn't a solution. Show me published control changes, reviews that are actually independent, and a consequence when someone falls short. Until then, this is a great day for communications, but unfortunately, not much of a day for safety.
That was pretty good. You got a notification. You might want to get that.
I know. You love when I type during our calls and you like it even more. Listen, if you don't let me tie, I should mute, but if you don't let me type, then how am I going to have AI tell me what to say?
I just pre-run it. I have literally, I have a, I have workflows that will automatically send it to my inbox. But sometimes you become
More from The Six Five with Patrick Moorhead and Daniel Newman
