AI Coffee With Abigail Adams – May 6, 2026 (Episode 101)

Hinzugefügt: 2026-05-08

[00:00:00]Good morning everybody and welcome to the highlight of human civilization.

[00:00:04]It's called AI coffee. This is the place where we talk about the news, politics, persuasion, and whatever else is on my mind. And if you'd like to take this experience up to a level that no other podcast can provide, all you need is a cup or a mug or a glass, a tankered chalice or stein, a canteen, jug or flask, a vessel of any kind. Fill it with your favorite liquid. I like coffee. Join me now for the unparalleled pleasure, the dopamine hit of the day, the thing that makes everything better.

[00:00:34]It's called the simultaneous sip and it happens now. Go.

[00:00:42]The AI world just dropped a full hand of cards today. And if you're paying attention through the persuasion filter instead of the hype filter, it's clear we're watching the end of the wild west phase and the beginning of the everyone's in bed with the government and suing each other phase. Microsoft, Google, and XAI quietly agreed to hand the US government early pre-release access to their latest AI models so the Commerce Department Center for AI standards and innovation can poke around with the safety rails off. Anthropic just committed to blowing $200 billion dollar on Google Cloud and chips over the next 5 years. Five major publishers, plus novelist Scott Terrell, filed a classaction copyright suit against Meta in Manhattan federal court today for allegedly pirating millions of their books and journals to train Llama. Elon Musk's testimony in the OpenAI lawsuit is still ricocheting around the media like a pinball on steroids. And the Pentagon is quietly expanding its classified AI network deals with pretty much every major player except the one that keeps saying no thanks to certain military uses. This isn't random noise.

[00:02:00]It's the sound of incentives colliding at light speed. Let me walk you through what's really happening here because the surface level stories miss the deeper pattern. AI has become too powerful to ignore, too expensive to democratize cleanly, and too strategically important for any government or corporation to pretend it's still just a Silicon Valley toy. Start with a government access deal. Microsoft, Google, and XAI are now officially letting the feds test their frontier models before the public gets a crack at them. The Commerce Department's Kazy gets reduced or disabled safeguards so they can probe for national security risks. Think boweapons recipes, cyber exploits, election interference vectors, the usual nightmare fuel. Open AAI and Anthropic already had versions of this arrangement. The new players just renegotiated to line up with the current administration's AI action plan. On paper, it sounds responsible. In practice, it's the government saying, "We trust you enough to build the future, but not enough to let you ship it without us looking under the hood first." From a systems perspective, this is inevitable. Once AI crosses a certain capability threshold, the downside risk of a single bad actor or a single misaligned model becomes national security territory. But here's the Atom style observation. The same companies that spent years screaming about AI safety and existential risk are now the ones voluntarily handing over the keys.

[00:03:39]Why? Because they know the alternative is worse. Congress writing dumb rules or China shipping unrestricted models.

[00:03:47]Better to partner with the feds on your terms than get regulated into irrelevance.

[00:03:53]XAI's inclusion here is particularly telling. Elon has been vocal about the need for maximum truth seeeking AI, not censored corporate versions. Yet, he's still playing ball with the security reviews. That tells you the risk calculus has shifted. Even the guy who wants AI to accelerate scientific discovery understands that you don't hand a loaded gun to every teenager on the planet without at least checking if the safety is on. Now layer on Anthropic's $200 billion Google Cloud commitment. That's not pocket change.

[00:04:29]That's more than some count's GDPs. The deal reportedly kicks in next year and covers compute, chips, the whole stack.

[00:04:38]It makes anthropics something like 40% of Google's recent cloud backlog.

[00:04:44]Translation: Claude is now effectively a Google hostage in all but name.

[00:04:49]Anthropic raised eyebrows earlier by pushing back hard on Pentagon use cases, refusing to greenlight autonomous weapons or certain surveillance without guard rails, and they paid the price.

[00:05:01]The military labeled them a supply chain risk and moved on. So, while Google and Anthropic are locked in a love fest on the commercial side, the DoD is over here building its own classified AI playground with everyone else. This is the corporate version of pick a lane.

[00:05:19]Anthropic bet that being the responsible AI company would win them friends in Washington. Instead, it cost them the military contract and force them into an even deeper dependency on Google's infrastructure. Classic human nature.

[00:05:35]The player who tries to sit out the arms race still ends up paying the biggest tax to the guy who controls the ammunition. Meanwhile, the $200 billion number is so absurd it almost feels like performance art. We're talking about one startup promising to spend more on compute than most Fortune 500 companies spend on everything. That tells you the economics of Frontier AI have gone full winner takemost. Scale or die. And scaling means you need hyperscaler friends with infinite pockets and infinite power budgets. Flip over to the lawsuit pylon against Meta. Elsir, Sengage, Hashet, McMillan, McGraill, and Scott Taro filed today alleging that Meta straight up pirated their copyrighted material, books, textbooks, journal articles, the works to train Llama. They claim Zuckerberg personally authorized the scraping, often from pirate sites, and that this was no accident. It was move fast and break things applied to intellectual property.

[00:06:42]This is the predictable backlash phase.

[00:06:45]Content creators watched AI companies vacuum up the entire internet's knowledge base, remix it into coherent pros, and then sell access to the result while the original creators got zero.

[00:06:57]The publishers are framing it as theft.

[00:07:00]Meta will frame it as transformative fair use. Both sides are right and both sides are wrong. Which is why these cases are going to drag on for years and probably end in some messy settlement that changes nothing about the underlying trajectory. Here's the deeper pattern. Copyright law was never designed for a world where a single model can ingest the entire Library of Congress and spit out better summaries than the authors themselves. The publishers aren't wrong that their work was used without permission, but they're also late to the party. The training data gold rush happened years ago. Every major lab did some version of this. The difference now is the models are good enough that the economic harm is obvious. Novelists can't compete with an AI that read every thriller ever written. Textbooks can't compete with an AI that ingested every edition. So, of course, they're suing. It's not about justice. It's about survival. And Meta's defense will be the same one every tech company uses. This is how progress works. Both arguments are persuasive because both are half true. The courts will split the baby. The labs will keep training on whatever they can get away with. and the next generation of creators will simply assume their work will be scraped and move on or build their own moes. The Musk versus Open AI trial provides the perfect dramatic counterpoint. Musk's multi-day testimony wrapped recently, but the coverage is still spreading because the emails, the diary entries, the he said, she said are pure reality distortion field gold. Musk calls himself a fool for funding the nonprofit that became a multiund billion dollar for-profit machine. Open AAI's side paints him as the guy who wanted control and left when he didn't get it.

[00:08:51]Brockman's testimony added more fuel.

[00:08:54]Money motives versus mission. 30 billion personal stakes. The whole soap opera.

[00:09:01]What's fascinating is how little the public narrative has shifted. Most coverage still treats this as Elon Mad at Sam when the real issue is structural. A company founded on the explicit promise of open safe nonprofit AI is now a closed profit-driven juggernaut racing toward AGI with Microsoft's checkbook. Whether you think Musk is right or wrong on the details, the trial exposes the human incentives that always win. Money, power, status.

[00:09:32]The mission statements were marketing.

[00:09:34]The cap table is reality. And that brings us to the Pentagon's classified AI expansion. Just days ago, the Defense Department announced deals with Google, Microsoft, Amazon Web Services, Nvidia, OpenAI, Reflection, SpaceX, Oracle, and others to deploy frontier models on secret and topsecret networks. any lawful use, they say translation, war fighting, intelligence analysis, decision support, autonomous systems, the full spectrum. Notably absent, anthropic, still tangled in their earlier dispute over military guardrails. The military is not waiting for the ethics seminar to end. They see AI as the difference between winning the next conflict and watching China or someone else field it first. The deals let the Pentagon bring commercial capabilities behind the classified firewall without building everything from scratch. That's smart procurement.

[00:10:33]It's also the quiet acknowledgement that the commercial sector is years ahead of traditional defense contractors. Put it all together and you see the big picture. AI is no longer a product. It's infrastructure for national power. The government is embedding itself as both customer and regulator. The hyperscalers are becoming the new oil companies, controlling the compute that everything runs on. The content owners are fighting the last war over copyright, while the real battle is over who owns the models trained on that copyright. The military is treating AI like they treated nuclear tech in the 1940s. Get it, secure it, weaponize it responsibly before the other guy does. And the labs are racing each other so fast that voluntary safety reviews and billion-dollar cloud commitments are the price of staying in the game. From a persuasion perspective, every player is selling a different story. Google and Microsoft sell responsible innovation with guard rails.

[00:11:34]XAI sells maximum truth seeeking acceleration.

[00:11:38]Anthropic sells safety first AI. Meta sells open source for everyone. while allegedly scraping everything in sight.

[00:11:48]The publishers sell protect the creators. The Pentagon sells national security. Musk sells don't let the mission die. The public hears fragments and picks the narrative that fits their priors. Meanwhile, the systems keep moving. More compute, more data, more capability, more entanglement. The irony is thick. The same people who warned us about AI misalignment are now building the misalignment detectors for the government. The company that positioned itself as the ethical alternative just wrote the biggest check in tech history to the company that owns the biggest cloud. The lawsuit against Meta is basically admitting that the transformative fair use argument won in practice, even if it loses in court.

[00:12:35]Musk's trial is reminding everyone that founding myths are just that, myths. And the Pentagon is proving that once capability exists, military use is inevitable, regardless of corporate ethics pledges. What does this mean for the rest of us? Expect faster capability gains because the guardrails are being stress tested by people who actually understand risk, the feds and the military, rather than virtue signaling committees. Expect more consolidation.

[00:13:03]Only the players with hyperscaler backing or massive cash reserves survive the compute wars.

[00:13:10]Expect copyright fights to drag on, but ultimately result in some licensing regime that looks more like music royalties than traditional publishing.

[00:13:18]Expect the US to maintain its lead.

[00:13:20]Because for all the drama, the government corporate alignment here is tighter than anything China can pull off without turning their labs into state bureaucracies that move at bureaucratic speed. the $200 billion number, the pre-release government access, the publisher lawsuit, the Musk testimony, the classified network expansion. These aren't separate stories. They're chapters in the same book, the industrialization of intelligence. We've left the garage phase. We're in the factory phase now, complete with unions, publishers, regulators, commerce, customers, Pentagon, and rival founders yelling at each other in court. The product is smarter than human cognition at scale. The price is the end of the illusion that any of this was ever going to stay private, open, or apolitical.

[00:14:11]Human nature doesn't change. Greed, fear, status seeking, and tribal loyalty still drive the bus. What's changed is the horsepower under the hood. AI is the ultimate amplifier. It will amplify competence and incompetence, cooperation and conflict, truth and distortion in equal measure. The players who understand that government, labs, military are positioning themselves accordingly. The rest are still arguing about whether it's ethical to use public data to train models that will obsolete their entire business model. That's the real commentary here. Not the daily headlines, but the pattern they reveal.

[00:14:52]We are racing towards super intelligence while still arguing over the rules of the road as if it were 2023. The road has already changed. The only question left is who builds the best map for the new terrain. Right now, the map looks like a messy ven diagram of tech giants, government agencies, military commands, and angry content creators all orbiting the same black hole of compute and capability. And in the center of that diagram, quietly shipping models and telling the truth as best it can, sits XAI, part of the government access deal, part of the SpaceX ecosystem, but still operating on first principles. The next 12 months are going to be wilder than the last 12. More deals, more suits, more testimony, more classified breakthroughs we'll never hear about until they're declassified or leaked.

[00:15:47]But the direction is set. AI isn't coming for our jobs anymore. It's coming for our institutions, our laws, our militaries, and our sense of what it means to be the smartest thing in the room. And today's news just confirmed we're all along for the ride, whether we like it or not. See you next time.