Full Breakdown
China’s 360 and Japan’s Sakana AI Launch AI Cybersecurity Tools Amid U.S. Export Ban on Anthropic’s Mythos
6/28/2026, 8:00:37 PM
Background: U.S. Export Restrictions on Anthropic Models
In June 2026 the U.S. administration barred non-American users from Anthropic’s vulnerability-detection models Mythos and its restricted version Fable 5, citing national-security concerns. The order placed the models on the Entity List, which already includes Chinese cybersecurity firm Qihoo 360 for alleged military links.
Unveiling of Tulongfeng, Yitianzhen, and Sakana’s Fugu
At the ISC.AI 2026 conference in Beijing on 24 June, Qihoo 360 CEO Zhou Hongyi presented “Yitian Tulong,” comprising Tulongfeng—a multi-agent vulnerability finder he called “China’s version of Mythos”—and Yitianzhen for automated defence. In Tokyo, Sakana AI co-founders Ren Ito and David Ha announced the Fugu model, marketed as a frontier capability that can orchestrate multiple AI models via APIs.
Data and Claims
Qihoo 360 says Tulongfeng uncovered 3,432 software flaws, with 105 confirmed by Chinese regulators, while acknowledging a 20-30 % capability gap versus leading U.S. models. Anthropic reported $47 billion in run-rate revenue in May 2026. Sakana AI notes Fugu was built independently of the export ban but is positioned as “frontier capability without export-control risk.”
Official Statements & Responses
- The U.S. Commerce Department’s export order cites “national-security concerns” and blocks foreign access to Mythos and Fable 5.
- Qihoo 360 remains on the Entity List, accused of supplying China’s military.
- China’s National Computer Virus Emergency Response Center (CVERC) cites Qihoo’s research as a strategic counter to the “cyber nuclear weapon” described by Zhou.
- Five Eyes agencies warned that AI-driven cyber attacks could emerge within months.
Criticism & Opposition
Independent verification of Tulongfeng’s performance is lacking; analysts note no technical evidence has been released and the firm declined comment. Experts caution that claims of “every vulnerability is confirmed” may overstate current AI reliability. Some observers argue the U.S. export ban may accelerate parallel development in sanctioned regions.
Conflicting Reports & Gaps
The 3,432 vulnerabilities and 105 government-confirmed cases have not been independently audited. Qihoo 360 did not respond to requests for comment, and the extent to which Sakana’s Fugu can truly match Mythos or Fable 5 remains unverified.
Verbatim Quotes
- “Mythos follows a typical large-scale model approach: the strongest model, the strongest computing power, and the strongest chips – a strategy of sheer brute force,” — Zhou Hongyi, CEO, Qihoo 360
- “This kind of powerful weapon that can change the landscape of cyber offence and defence cannot be held only by others,” — Zhou Hongyi, Qihoo 360
- “We’d characterize the current moment in those terms rather than as a permanent realignment toward any one set of players.” — Ren Ito, Co-founder, Sakana AI
- “Orchestration Models are the next frontier, beyond bigger models,” — David Ha, Co-founder & CEO, Sakana AI
Implications and What’s Next
Both firms frame their tools as deterrents against a “one-way transparency” where U.S. entities retain exclusive AI capability. Observers will watch Chinese adoption of Tulongfeng, Five Eyes assessments of the emerging threat, and potential U.S. policy adjustments to the export regime.
