APREX · EST. 2026
When Anthropic unveiled its new AI model called Mythos in April 2026, few could have predicted it would trigger one of the most dramatic confrontations between an AI company and the US government in history. Two months later, the model had been globally suspended, reinstated, and placed at the center of a broader debate about who should ultimately control the most powerful AI systems now in existence.
A Model That Found Zero-Day Vulnerabilities in Hours
Mythos 5 was introduced on April 7, 2026, and is described by Anthropic as being built with particularly strong capabilities in agentic coding and advanced reasoning. During pre-release testing — and as part of a collaboration with US intelligence agencies known as "Project Glasswing" — the model demonstrated the ability to identify thousands of previously unknown zero-day vulnerabilities across all major operating systems and browsers, according to research cited by MIT Technology Review.
Even more alarming to authorities: Mythos 5 reportedly uncovered vulnerabilities in classified US government systems within a matter of hours. Anthropic itself acknowledged that AI models have now reached a level where they can surpass all but the most exceptionally skilled human hackers.
Anthropic acknowledged that AI models have now reached a level where they can surpass all but the most experienced human hackers.
A Jailbreak Triggered Export Controls
What caused the immediate crisis to erupt was not Mythos alone, but the revelation that Amazon researchers had managed to circumvent the safety guardrails of the related model Fable 5. According to MIT Technology Review, Amazon CEO Andy Jassy alerted Trump administration officials that these guardrails had serious weaknesses.
On June 12–13, 2026, Commerce Secretary Howard Lutnick intervened with an export control directive that barred foreign nationals — including Anthropic's own employees of foreign origin — from accessing Fable 5 and Mythos 5. Anthropic responded by globally suspending both models.
Security Experts Pushed Back
The response from the professional community was swift. Dozens of prominent cybersecurity experts sent an open letter to Commerce Secretary Lutnick and National Cyber Director Sean Cairncross, characterizing the directive as "dangerous." They argued that the ban denied defenders access to the best available tools, created market uncertainty, and undermined US leadership in AI — without the actual risk justifying the intervention, according to MIT Technology Review.
Anthropic's own position was that the jailbreak technique that triggered the crisis had only exposed "minor" security flaws that other publicly available AI models could also find. The company warned that such an industry standard would in practice "halt all new model rollouts for all frontier providers."
Three Things to Watch Going Forward
MIT Technology Review identifies three key areas that will shape the further development of this conflict:
1. Anthropic's Red Lines
Anthropic has established clear prohibitions against its models being used for military and surveillance purposes. These principles formed part of the backdrop for the Pentagon's earlier designation of the company as a national security risk, and it remains unclear how these guidelines will hold up under sustained government pressure.
2. The Precedent for AI Export Controls
The directive against Anthropic marks a significant departure from the Trump administration's previously hands-off approach to AI regulation. If Fable 5 remains blocked without a clear rationale, it could set a precedent for far more interventionist government control over which models are permitted to be released at all.
3. The Balance Between Security and Innovation
The interim resolution — in which Mythos 5 is now being released to more than 100 American institutions, including major corporations and government agencies — was described by MIT Technology Review as the result of "significant progress" achieved through daily negotiations. Anthropic has committed to working on protocols and standards for its models. The question is whether this represents a lasting compromise or merely a temporary pause in a more structural tension.
What Happens Now?
As of June 28, 2026, Mythos 5 is back in operation for a limited number of American actors, while Fable 5 remains under restrictions. The conflict has made clear that the US government is willing to use export controls as a tool against AI companies when it judges national security interests to be at risk — and that this can happen quickly and with limited advance warning.
For the industry as a whole, the episode raises a fundamental question: when AI models reach a point where they can independently uncover zero-day vulnerabilities in government systems, who is ultimately responsible for drawing the lines?
Sources: MIT Technology Review (June 22, 2026)