Ian Bremmer says Anthropic's new AI model is dangerous to all of us globally

Ian Bremmer did not come on The Diary of a CEO to do the usual AI podcast séance, half awe, half LinkedIn panic. He made a specific accusation: Anthropic’s newest model, he said, is dangerous to everyone because it can find exploitable software bugs.

That is the kind of sentence that makes Silicon Valley executives suddenly discover the phrase “context window.” Bremmer is not a random doom merchant with a Substack and a ring light. Ian Bremmer is the founder of Eurasia Group, a political scientist whose whole brand is telling governments and corporations which volcano is about to erupt. So when he frames an AI model less like a product update and more like unsecured fissile material, the interesting part is not whether he sounds dramatic. He does. The question is whether the drama is earned.

That model is dangerous to all of us globally because any software with a potential bug in it is findable by that model and it can be exploited.

The claim is simple enough to be scary and vague enough to be very useful on a stage. Bremmer does not name the model in this clip, and he does not give a worked example of the exploit chain he has in mind. But the underlying fear is real. Modern AI systems are getting better at code analysis, vulnerability discovery, and automation. That does not mean every powerful model is a cyber weapon. It does mean the line between “helpful security assistant” and “excellent intern for criminals” is getting thinner.

Bremmer wants an AI Financial Stability Board

Bremmer’s fix is not the usual podcast prescription of “be mindful” or “teach your kids to code, but emotionally.” He wants institutions. Boring, acronym-friendly, grown-up institutions. His model is the financial system, where central banks and regulators coordinate during crises because nobody benefits if the whole casino burns down, not even the people holding the chips.

You need AI stability board, so that whenever there is a model that creates a danger to us globally, like Anthropic just did last week.

This is Bremmer at his most Bremmerian: take a messy technology story and drag it into geopolitics, where everything becomes a systems risk with flags attached. He wants an independent body with enough technical ability to identify threats, warn people with power, and respond quickly. That sounds sensible until you ask the ugly questions. Who appoints this board? Who trusts it? What happens when a U.S. company, a Chinese lab, and an open-source developer all disagree about whether a model is too dangerous to release?

Still, his instinct is better than the current vibe, which is basically five companies racing toward superintelligence while everyone else refreshes press releases. Artificial intelligence is being governed by a strange cocktail of voluntary safety pledges, national security panic, investor pressure, and vibes. Bremmer’s point is that this is not enough. On that, he is right.

The nuclear analogy is useful, until it starts showing off

Bremmer also wants the United States and China to start AI arms control talks. Not eventually. Now. He compares the moment to the Cold War before the U.S. and Soviet Union learned, after nearly ending civilization during the Cuban Missile Crisis, that hotlines and limits were not sentimental niceties. They were plumbing for survival.

We desperately need that between the Americans and the Chinese.

The nuclear analogy is powerful because it forces a little humility into a business culture that treats every risk as a hiring problem. It is also imperfect. Nuclear weapons are hard to build, easy to count, and tied to states. AI is messy, distributed, copyable, commercial, open in places, secret in others, and already inside daily life. Arms control for missiles is hard. Arms control for model weights, chips, data centers, cyber tools, and research talent is a migraine wearing a lanyard.

But Bremmer’s larger AI governance argument is not just about China, Anthropic, or Sam Altman charging for intelligence like water. It is about the gap between people who can command AI and people who cannot even reliably access electricity. That is where his warning gets less techno-thriller and more morally bracing.

The gap between people with electricity and people don’t have electricity is going to be a hell of a lot worse when it’s AI.

This is the stronger version of his case. The Anthropic line will get the screenshots, because naming a company always does. But the real stake is access to power, literal and cognitive. If AI becomes the layer through which education, health care, government services, software, translation, finance, and work all move, then being locked out of it will not look like missing the latest app. It will look like being seated at the children’s table of civilization.

So yes, Bremmer is overselling when he turns one unnamed Anthropic release into a globally dangerous weapon without showing the receipts. He is not wrong to be alarmed. The tech industry has spent years asking the public to admire capability first and trust safety later. Bremmer is asking a ruder question: what if the adults are not actually in the room?