Cybersecurity Experts Push Back on US Ban of Anthropic’s Best Models

A bunch of cybersecurity veterans just sent a strongly worded letter to the White House, and it’s not about some new zero-day or ransomware gang. They’re pushing back on the US government’s decision to slap export control restrictions on Anthropic’s two most powerful models, Fable and Mythos.

Their argument is straightforward: these restrictions are going to make it harder for security teams to defend their own software and products. And when you hear who’s signing onto this thing—people who’ve spent decades on the front lines of cyber defense—it’s hard to dismiss as just another industry whine.

The government’s logic, I assume, is the usual one: keep advanced AI out of the hands of adversaries. But the cybersecurity crowd is calling this move “dangerous” because it kneecaps the very people trying to patch vulnerabilities and detect threats before they become headlines.

Anthropic has been positioning Fable and Mythos as serious tools for code analysis and threat detection. They’re not just chatbots that write poetry. These models can parse complex codebases, spot logic errors, and even suggest hardened configurations. That’s exactly the kind of capability defenders need when they’re drowning in alerts and short on time.

Banning exports of these models means teams outside the US—and even some inside, depending on how the restrictions are enforced—lose access to that horsepower. Meanwhile, offensive actors who don’t care about export controls will just use whatever they can grab from other sources. It’s the classic asymmetry problem: defenders follow the rules, attackers don’t.

The letter isn’t a full-throated endorsement of Anthropic either. These experts aren’t saying the models are perfect or that there’s no risk. They’re saying the risk calculus is off. Restricting access to advanced AI for defensive purposes while offensive capabilities proliferate elsewhere is a net loss for security.

I’ve seen this pattern before with encryption tools back in the 90s. Governments tried to limit strong crypto exports, claiming national security. What actually happened was that security products outside the US were weaker, and the bad guys still found ways to get strong encryption. The same dynamic is playing out here, just with neural networks instead of RSA keys.

What’s interesting is that this pushback is coming from people who usually lean toward caution on AI. They’re not AI hype bros. They’re the ones who’ve cleaned up after ransomware attacks and traced APT campaigns. When they say a policy is dangerous, it’s worth listening.

The White House hasn’t responded publicly yet, but this kind of pressure from the cybersecurity community carries weight. If the restrictions stay, we’ll likely see more workarounds and less secure software globally. If they’re loosened, the government will need to find other ways to manage the risks of models like Fable and Mythos falling into the wrong hands.

Either way, the debate is shifting from “should we restrict AI” to “who gets hurt most by the restrictions.” And right now, the defenders are saying they’re the ones taking the hit.