Anthropic's Mythos 5 LLM had everyone buzzing. Even CEO Dario Amodei admitted — or boasted, depending on your perspective — that the model would bring an "enormous increase in the amount of vulnerabilities, in the amount of breaches" to the world. But with that fear came promises of unprecedented AI capability. Then Washington stepped in and yanked the rug out.

The February Standoff

This isn't the first clash between Donald Trump's administration and Anthropic. Back in February, Anthropic refused to give Defense Secretary Pete Hegseth the power to use its models for spying on American citizens and powering autonomous weapons. That refusal apparently didn't sit well with the White House, setting the stage for this latest escalation.

Amazon's Not-So-Neutral Tip

Here's where it gets interesting. According to The Wall Street Journal, Commerce Secretary Howard Lutnick and Undersecretary William Kimmitt — neither of whom has any meaningful AI expertise — decided Fable 5 was a national security threat. But who tipped them off? Amazon CEO Andy Jassy. Guess what AWS offers as part of its cloud portfolio? Full AI infrastructure through Amazon Bedrock, SageMaker, and agentic AI tools. Amazon is not a neutral party in this fight — it's a direct competitor that also happens to have partnerships with Anthropic. The big players are all in bed with each other while still going for each other's throats.

The 90-Minute Ultimatum

Commerce gave Anthropic exactly 90 minutes to fix its "problem." AI development moves fast, but not that fast. Officials never spelled out specifically what was wrong — only referencing a "narrow, non-universal jailbreak" in Fable. That's it. That's all. Meanwhile, similar jailbreaks exist in OpenAI's GPT-5.5, yet ChatGPT maker hasn't faced comparable export controls. The inconsistency is glaring.

Industry Pushback

Security and AI experts aren't buying the official rationale. In an open letter titled "On Transparent AI Cyber Protections," they wrote that Commerce's directive "has taken the best models away from defenders, created market uncertainty, and risked America's AI leadership without any real risk to justify it." Alex Zenla, co-founder and CTO of security company Edera, pointed out that Fable's capability to identify insecure code sections is baseline functionality — present in GPT-5.5, Opus, Sonnet, and Kimi 2.7. "Pulling capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous," the experts warned.

The Real Motivation

This isn't about AI safety or security. It's about Trump opening another front in his war against Anthropic — a company that hasn't groveled to the administration like other tech firms have. But here's the irony: by essentially killing Anthropic's biggest move, Trump's team is telling the world you can't rely on American AI companies as stable partners. In the long run, expect Europe and other nations serious about digital sovereignty to think twice before trusting any US-based AI provider.

Negotiations in Progress

Anthropic and Commerce officials are now locked in tense talks over whether — and under what safeguards — Fable 5 and Mythos 5 could return. Commerce suggests a narrower relaunch might be possible if jailbreak issues resolve and additional controls exist. What those controls look like remains unclear, but given the administration lacks anyone with actual AI chops calling the shots, don't expect rational policy.

Key Takeaways

  • Anthropic's refusal to enable mass surveillance for Defense Secretary Hegseth appears connected to Washington's vendetta
  • Amazon flagged Fable 5 despite being a direct competitor with its own AI stack
  • Similar vulnerabilities exist in GPT-5.5, but OpenAI hasn't faced equivalent restrictions — the double standard is obvious
  • Experts warn pulling capabilities from defenders while adversaries advance is actively harmful

The Bottom Line

Trump's Commerce Department just told the world American AI companies can be weaponized for political purposes on a whim. That's not how you maintain tech leadership — that's how you hand market share to Beijing and Brussels.