May 12, 2026 · 30 min · 13 segments
Sam Altman’s credibility just took a hit. In Episode 77 of What the AI?!, Jeff Keltner and Annie Delgado break down a chaotic week where legal drama, national security, and real-world AI performance…
Annie DelgadoHost
I've been, like, trying to keep track of, of how this is all unfolding, so let me tell you what I understand, and you can tell me if I've got my facts right, at least, at least as you understand them.
So the supply chain risk, I think the best argument for that is, like, uh, the government says it's a supply chain risk if we use Anthropic because we're gonna become so dependent on them as our model provider, but there's these restrictions in how we're allowed to, to use it.
This is our only tool to do this thing, but we can't do it because we have this, you know, you know, restriction from them, then that puts us at risk, and so therefore it's a supply chain risk.
They're worried about sort of getting too, too hooked on it and having a guardrail.
But then they launched Mythos, and the government saw how amazing Mythos is and went, "Whoa, whoa, whoa, we can't...
We can't just have this in the hands of anybody, for we first need it in the hands of ourselves, and then we need to control it." Is that sort of the basics around-
That's, that's sort of the basics, and with Mythos, they actually went to Anthropic i- in the midst of this supply chain debate and said, "No, you can't expand this access to more people," not so much, my understanding is, out of concerns about people having access to the tool, but out of concerns that there wouldn't be enough compute left for the government to get as much access as they wanted to it.
So they're kind of like, "No, no, we need more and more of that thing even though we're..." You know.
They're not worried that the models are somehow, uh, you know, infused with dangerous capabilities, vis-à-vis China.
They're really concerned about Anthropic's guardrails, and so that's a separate concern from saying, "Hey, the model as it exists and the things we wanna do with it, your model is really great." Um, so it does seem like, I think, again, we've said this before, the wrong thing.
But to add, you know, some in, in, some more layers to this, this week, so this, uh, the DoD came out, the Pentagon, and awarded, um, classified network access to eight AI companies, SpaceX with their OpenAI, uh, sorry, with xAI, OpenAI, Google, NVIDIA, Reflection, Microsoft, AWS, Oracle, and notably missing from that list is Anthropic.
Um, and they are now ac- giving all those companies access to classified networks, um, or allowing their AI models to access classified networks.
It doesn't seem like a coherent response from the government, at least as of yet.
It, it sure doesn't, and I think we're gonna see more unfold about this because, you know, we're gonna see this executive order potentially and, and so on and so forth.
But it really sort of begs the question of, like, how good does your product have to be that the government is basically like, "It's so good that everybody else is gonna use it, and they're gonna take all my tokens"? [laughs] And like that's...
Even the government is worried about its tokens and how it's gonna be able to use the, to use Anthropic, so it's, it's kind of crazy to think about just how incredible their technology is.
I've been, like, trying to keep track of, of how this is all unfolding, so let me tell you what I understand, and you can tell me if I've got my facts right, at least, at least as you understand them.
So the supply chain risk, I think the best argument for that is, like, uh, the government says it's a supply chain risk if we use Anthropic because we're gonna become so dependent on them as our model provider, but there's these restrictions in how we're allowed to, to use it.
This is our only tool to do this thing, but we can't do it because we have this, you know, you know, restriction from them, then that puts us at risk, and so therefore it's a supply chain risk.
They're worried about sort of getting too, too hooked on it and having a guardrail.
But then they launched Mythos, and the government saw how amazing Mythos is and went, "Whoa, whoa, whoa, we can't...
We can't just have this in the hands of anybody, for we first need it in the hands of ourselves, and then we need to control it." Is that sort of the basics around-
That's, that's sort of the basics, and with Mythos, they actually went to Anthropic i- in the midst of this supply chain debate and said, "No, you can't expand this access to more people," not so much, my understanding is, out of concerns about people having access to the tool, but out of concerns that there wouldn't be enough compute left for the government to get as much access as they wanted to it.
So they're kind of like, "No, no, we need more and more of that thing even though we're..." You know.
They're not worried that the models are somehow, uh, you know, infused with dangerous capabilities, vis-à-vis China.
They're really concerned about Anthropic's guardrails, and so that's a separate concern from saying, "Hey, the model as it exists and the things we wanna do with it, your model is really great." Um, so it does seem like, I think, again, we've said this before, the wrong thing.
But to add, you know, some in, in, some more layers to this, this week, so this, uh, the DoD came out, the Pentagon, and awarded, um, classified network access to eight AI companies, SpaceX with their OpenAI, uh, sorry, with xAI, OpenAI, Google, NVIDIA, Reflection, Microsoft, AWS, Oracle, and notably missing from that list is Anthropic.
Um, and they are now ac- giving all those companies access to classified networks, um, or allowing their AI models to access classified networks.
It doesn't seem like a coherent response from the government, at least as of yet.
It, it sure doesn't, and I think we're gonna see more unfold about this because, you know, we're gonna see this executive order potentially and, and so on and so forth.
But it really sort of begs the question of, like, how good does your product have to be that the government is basically like, "It's so good that everybody else is gonna use it, and they're gonna take all my tokens"? [laughs] And like that's...
Even the government is worried about its tokens and how it's gonna be able to use the, to use Anthropic, so it's, it's kind of crazy to think about just how incredible their technology is.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.