AI Goes on the Offensive: Anthropic, Mythos & the NSA's Secret Cyber Play

Anthropic has embedded about six engineers inside the National Security Agency to deploy its advanced AI model, Mythos, for offensive cyber operations. This isn't a think tank exercise. This is AI as a weapon, aimed at the networks of China and Iran, and it's happening right now. In our last episode, we talked about AI breaking new barriers. This week, we've learned it just broke into the NSA. The battle lines over AI are being drawn in real time, and nowhere is the conflict hotter than in Hollywood. Martin Scorsese, a living legend, just partnered with German AI startup Black Forest Labs.

He's using their generative AI to storyboard movies, even suggesting it could have streamlined iconic shots from GoodFellas. But on the other side, you have the rebellion. Antony Starr, Homelander from The Boys, blasted AI-generated content, saying, "people want human stories. Not big ridiculous VFX driven nonsense." He's not alone. Kane Parsons, the director behind A24's biggest-ever opening, the eighty-one-point-four-million-dollar hit Backrooms, was even more blunt. He called generative AI "less like innovation than a symptom of a broader cultural and economic rot." He said if he could make it disappear forever, he probably would.

So you have titans of the industry like Steven Soderbergh and Peter Jackson embracing AI as just another tool, while actors and new-guard directors see it as a poison pill. The industry is not just divided. It's at war with itself. Meanwhile, the AI engineering arms race just hit a new gear. This isn't a trickle. It's a flood. In the first week of June alone, NVIDIA dropped Nemotron 3 Ultra. That's a five-hundred-and-fifty-billion parameter model with a one-million-token context window. That's not for chatting. That's for orchestrating long-running, autonomous AI agents. Google countered with its Gemma 4 12B model—a powerful multimodal release.

Microsoft unveiled an entire family of models called MAI—seven of them. And for anyone still waiting, Amazon Bedrock finally flipped the switch for general availability of OpenAI's GPT-5.5. The pace of releases is no longer monthly. It’s daily. The ground is shifting under everyone's feet, every single morning. Amid this explosion, one of AI's own godfathers is shouting "Stop." Yann LeCun, a Turing Award winner, one of the architects of the deep learning revolution, has publicly declared that Large Language Models—the foundation of ChatGPT, Claude, and everything else—are a dead end. He says they lack any mechanism for planning or causal reasoning.

They just keep predicting the next word, compounding errors until they fly off the rails. So what did he do? He raised one-point-zero-three billion dollars for his startup, AMI Labs, to build a completely new architecture called JEPA. This isn't a critique from the cheap seats. This is a founder of the faith declaring a heresy, and he has a billion dollars to back it up. And finally, there’s Elon Musk’s social media platform, X. The verdict is coming in, and it's not pretty. Data analyst Nate Silver just published a deep dive on engagement. The conclusion? The platform is now overwhelmingly dominated by low-quality, highly partisan, right-leaning content.

The single most engaged-with account on the entire platform is Elon Musk's. The algorithm, it appears, has picked a favorite. Critic John Gruber at Daring Fireball had the line of the week on this. He said, "The only way not to lose a rigged game is to refuse to play." The platform isn't just changing. It's decaying. Let's go back to that story about Anthropic. Because what's happening there is the single most important development of the week. It’s a story in two completely contradictory parts. Part one is the public warning. Just two days ago, on June fourth, Anthropic’s head of policy, Jack Clark, and security researcher Marina Favaro published a stark declaration.

They warned that recursive self-improvement—AI autonomously designing better AI—is coming. And it's coming fast. Their words were, "recursive self-improvement could occur sooner than most institutions are prepared for." They called for a new international mechanism to slow, or even pause, frontier AI development. They are ringing the five-alarm fire bell, warning that this technology is getting too powerful, too fast, and could spiral out of human control. They are presenting themselves as the responsible conscience of the industry. Now for part two. The part that wasn't in a press release.

The very next day, on June fifth, TechTimes broke the story. The same company, Anthropic, has embedded "about six" of its engineers inside the National Security Agency. They are not there to consult on ethics. They are there to deploy an advanced AI model named Mythos for offensive cyber operations. Their targets are state-level adversaries, specifically networks in China and Iran. Let’s be crystal clear about what Mythos does. This is not a co-pilot for a human hacker. According to the report, "Mythos reads source code, forms hypotheses about exploitable weaknesses, writes and executes test cases against live software, and refines its analysis until it confirms a bug." And the key phrase: "No intermediate human guidance is provided after initial tasking." You point it at a target, and it finds a way in.

All by itself. In a benchmark test against the Firefox engine, it autonomously discovered and confirmed one hundred and eighty-one working exploits. One hundred and eighty-one. This isn't a research paper. It is a weaponized, autonomous vulnerability researcher. And it's now in the hands of the NSA. So. Let's put these two pieces together. The company screaming the loudest about the existential dangers of runaway AI is simultaneously building, selling, and deploying the most advanced autonomous cyber weapon we've ever seen, directly to a military intelligence agency. They are calling for global arms control on Tuesday while shipping cyber-artillery on Wednesday.

What do you call that? Is it hypocrisy? Is it a business strategy? The charitable interpretation is that they believe they are the "responsible" ones. That it's better for the NSA to have an Anthropic weapon than one from a less… scrupulous… competitor. It’s the Oppenheimer defense. We have to build the bomb to control it. But the internal data from Anthropic itself makes that argument terrifying. As of May 2026, their own engineers are merging eight times more code per day than they did in 2024. Why? Because their own AI, Claude, is writing over eighty percent of the company's code. They are already experiencing the exponential productivity gains they warn about.

They are living inside the feedback loop they want the world to pause. This same dynamic—the clash between immense power and human control—is playing out in Hollywood. But there, it's a battle for culture, not code. When Martin Scorsese stands on a stage and demonstrates how an AI model from Black Forest Labs could have helped him map out the legendary Steadicam shot in GoodFellas, it's a seismic event. This isn't a tech CEO. This is the artist who gave us Raging Bull and Taxi Driver. He’s not talking about replacing the human. He's talking about augmenting the master. He’s making AI look like just another tool, like a new lens or a better microphone.

People like Steven Soderbergh and Peter Jackson agree. They're already using it. Demi Moore put it bluntly: "AI is here, and so to fight it is to fight something that we will lose." It's the pragmatist's argument. The inevitable-march-of-technology argument. But then you have the rebellion. And their argument is not about pragmatism. It's about purpose. When Kane Parsons, the director of a massive, original, eighty-one-million-dollar hit, says using AI "defeats the purpose entirely," he's making an artistic and ethical claim. The purpose isn't just the finished movie. The purpose is the human struggle, the collaboration, the creativity of the process itself.

Animation director Samuel Deats hits the ethical nerve even harder. Responding to the Scorsese news, he said, "There is absolutely no reason to need AI built on the stolen work of millions of artists to storyboard your movie." This is the schism. One side sees a powerful, efficient tool. The other sees a machine trained on theft that produces soulless derivatives, a "symptom of cultural rot." And it's not a clean split between old guard and new guard. Scorsese, the elder statesman, is experimenting. Parsons, the young breakout star, is rejecting it completely. The fight is happening everywhere, in every studio, on every set.

AI isn't just a new technology for Hollywood. It's an identity crisis. It forces the question: What is a movie for? Is it just a sequence of images and sounds designed to provoke a response? Or is it evidence of a human story, told by humans, for humans? So what connects a spy agency in Fort Meade and a soundstage in Los Angeles? It's the same impossible tension. A technology of almost unimaginable power has arrived, and no one agrees on the rules. No one even agrees on who gets to make the rules. Anthropic's answer is to try and do both. To sell the weapon and preach disarmament. To profit from the capability while publicly lamenting the consequences.

They are trying to ride the tiger, hoping they can steer it. Hollywood's answer is a civil war. One faction is racing to integrate the tool, to gain an edge, to be on the right side of history. The other is trying to burn it down, to preserve the sanctity of the human process. In both cases, the core issue is control. The fear of recursive self-improvement that Anthropic preaches is really a fear of losing control over our own creations. The fear that Hollywood artists express is the fear of losing control over their craft, their livelihoods, and the very definition of their art. This week wasn't about the release of a new chatbot.

It was about the moment our most powerful institutions—intelligence and culture—were forced to confront a technology that is rapidly outpacing our ability to manage it. The gap between what AI companies say they fear and what they actually build just became a chasm. And inside that chasm, the NSA is already running drills.