Key Points
- On June 13, 2026, the Trump administration forced Anthropic to terminate global access to its Mythos and Fable 5 AI systems
- According to Semafor, intelligence suggests a Chinese-affiliated entity may have obtained unauthorized access to Mythos
- American security officials worry China could replicate the model’s capabilities using a technique known as “distillation”
- Reports emerged that Fable 5, the consumer-facing iteration of Mythos, had been successfully jailbroken — though Anthropic challenged these claims
- Cryptocurrency exchanges like Coinbase were cut off from critical AI defense systems provided through Project Glasswing
In a dramatic development, Anthropic terminated international access to a pair of its artificial intelligence systems on June 13, 2026. The shutdown followed explicit instructions from the Commerce Department, which invoked national security justifications.
The affected systems were Mythos and Fable 5. According to Anthropic’s statement, the company lacked the technical infrastructure to authenticate user nationality during individual API interactions, prompting the decision to eliminate access completely rather than pursue selective restrictions.
According to exclusive reporting by Semafor journalist Reed Albergotti, the termination was partially motivated by intelligence indicating a China-connected organization had infiltrated Mythos.
American intelligence agencies expressed concern that Chinese possession of the model could enable replication of its functions via “distillation” — a technique where a secondary AI system learns by analyzing outputs from a more sophisticated model.
Anthropic clarified that White House officials never mentioned Chinese infiltration concerns during direct communications. Instead, those conversations centered on Fable 5 and allegations regarding its security compromise.
Mythos’ Capabilities Explained
Introduced in April 2026, Mythos represents a cutting-edge AI engineered specifically for cybersecurity applications. Anthropic determined the technology posed too significant a risk for unrestricted public distribution.
The system operates autonomously to discover and weaponize previously unknown software security flaws. Documentation from Anthropic reveals it uncovered thousands of critical zero-day vulnerabilities spanning prominent operating systems and web browsers.
Notably, the model discovered a 27-year-old remote crash exploit in OpenBSD and demonstrated 83.1% success rates on the CyberGym vulnerability assessment. Anthropic’s analysis suggested it could condense months of specialized penetration testing into mere minutes.
Distribution was restricted to approximately 12 designated corporate allies and more than 40 institutions participating in Project Glasswing, a defensive security initiative.
This isn’t the inaugural security lapse. Reports indicate a Discord community maintained illicit Mythos access for a fortnight before Anthropic detected and terminated the connection.
The Fable 5 Security Breach and Cryptocurrency Consequences
Fable 5 represents a publicly available derivative of Mythos incorporating additional safety mechanisms. White House AI strategist David Sacks claimed on X that Anthropic’s CEO Dario Amodei received briefings about the jailbreak but minimized its significance.
Anthropic countered these assertions, characterizing the jailbreak as limited in scope and involving vulnerabilities already present in competing public models.
Sacks described the circumstances as “serious but easily resolved” and indicated responsibility rested with Anthropic to address the matter.
The access termination carries substantial implications for cryptocurrency and blockchain security operations. Coinbase had reportedly initiated negotiations to obtain Mythos access through Project Glasswing for protective security applications.
Prior to the prohibition, Zcash’s creator asserted that Mythos had conducted comprehensive audits of the complete Zcash blockchain without identifying vulnerabilities.
Security operations that depended on Mythos for vulnerability detection now face complete access denial. Anthropic has provided no timeline regarding potential restoration of availability.





