Anthropic just can’t catch a break. The AI company, already juggling lawsuits against the Pentagon and a rocky relationship with the US government, now finds itself in yet another high-stakes showdown. This time, it’s over a tool so powerful the company itself called it “too powerful” — and then had to pull it from public access entirely.
According to BBC reporting, Anthropic executives are set to meet with Secretary of Commerce Howard Lutnick in Washington DC on Monday. The agenda? Fresh national security concerns over the company’s latest release, a new AI model called Claude Mythos.
Here’s where things get messy. Anthropic actually released two versions. Fable 5 is the one with extra safeguards made available to the public. Mythos 5, with different controls, was only supposed to go to a select group of organizations. But within days of the public release, the US government essentially shut it down, prohibiting Anthropic from allowing any foreign national access to the technology. Anthropic responded by blocking all public access on Friday.
The weird part? The government says it “became aware” of a potential jailbreak vulnerability — essentially an opening for someone to make the AI do something it wasn’t designed to do. But Anthropic says it only received “verbal evidence” of this vulnerability. That’s a far cry from concrete proof.
So now we’re in this strange situation where the government is waving a red flag, the company is scratching its head trying to understand what’s actually happening, and everyone is meeting on Monday to figure out the next steps.
This isn’t exactly new territory for Anthropic. Earlier this year, the company sued the US Department of Defense over disagreements about how its models can be used. Tensions appeared to be cooling after what was described as a “productive” meeting with White House officials a few weeks ago, but this latest drama has reignited the tension.
What strikes me here is the asymmetry. We have a company that built something powerful enough that its own creators hesitated. Then we have a government reacting with alarm, but apparently without providing hard evidence of what exactly they’re worried about. Somewhere in the middle is an AI tool that may or may not have a genuine vulnerability — and the public is left in the dark.
The real question isn’t just about this particular model. It’s about what happens when AI companies and governments clash over technology that’s moving faster than either side’s ability to regulate or understand it. These meetings behind closed doors might resolve things temporarily, but they don’t build trust.
Whether Anthropic gets to release Fable 5 and Mythos 5 again after Monday’s meeting remains genuinely unclear. But one thing’s for sure: the honeymoon period for frontier AI companies operating in a regulatory vacuum is well and truly over.
