Qihoo 360, a Chinese cybersecurity firm on a U.S. blacklist, unveiled an AI tool for automated software vulnerability discovery called Tulongfeng at the ISC.AI 2026 conference held in Beijing around June 24, 2026, claiming it rivals the vulnerability-finding capabilities of Anthropic's unreleased Claude Mythos model.
June 24, 2026 · ISC.AI 2026 · Beijing
China's Qihoo 360 unveils "Tulongfeng," a bug-hunting AI it claims rivals Anthropic's restricted Mythos
Founder Zhou Hongyi framed the multi-agent tool as a "Chinese version" of Claude Mythos — a model Anthropic deemed too dangerous to release — and called such capabilities a "cyber nuclear weapon."
3,432
Vulnerabilities Tulongfeng claims to have found in open-source & commercial software
105
Of those, confirmed by Chinese authorities
0
Independent verifications — claims remain unconfirmed
Claim vs. claim
Two cyber-offense AIs, two very different postures.
Tulongfeng
Qihoo 360
Type · Multi-agent bug-finding swarm
Status · Said to be already operational
Findings · 3,432 (105 confirmed)
Framing · Self-styled "Chinese version" of Mythos
Claude Mythos
Anthropic
Type · Restricted frontier model
Status · Limited access via Project Glasswing
Findings · Thousands of zero-days (OS, browsers)
Framing · "Too dangerous to release"
Inside the "agent route"
A team of specialized agents, not a single LLM — built on two decades of threat data.
Threat
→
Data-flow
→
Sandbox &
Read as deterrence
Seen as China's bid for its own "AI deterrence" — a strategic asset to counter and balance a restricted frontier capability.
Read as hype
Critics flag the "cyber nuclear weapon" phrasing, possible imitation or distillation of Anthropic's model, and the lack of any independent test or user report.
The catch
No accuracy, false-positive, or real-world exploitability data has been independently confirmed.
Coverage is news-driven, with almost no reports from actual users — and Qihoo 360 sits on a U.S. blacklist, sharpening U.S.–China tensions over AI and cyber. Real assessment awaits public benchmarks and independent testing.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…