Qihoo 360 Says Tulongfeng Nearly Matches Anthropic’s Mythos
Summary
Qihoo 360 CEO Zhou Hongyi says its Tulongfeng system nearly matches Anthropic’s Mythos and has found 3,000+ vulnerabilities. The capability claim is unverified; the article frames AI vulnerability discovery as a dual-use arms race.
Key points
- Qihoo 360 CEO Zhou Hongyi announced Tulongfeng and said it is nearly capable of matching Anthropic’s Mythos.
- Zhou claimed Tulongfeng had found more than 3,000 vulnerabilities, many classified as high risk; the article says its capabilities have not been independently verified.
- The article says Mythos accelerated vulnerability discovery and exploit chaining across IT and OT systems, finding more than 10 times as many vulnerabilities as previous frontier models.
- Zhou compared AI-driven vulnerability discovery to a nuclear arms race and described the capability as a potential strategic deterrent.
- The article says the US restricted Anthropic’s model release over concerns about its capabilities, and that Anthropic later released Mythos 5, which it says surpassed Mythos Preview.
- The article describes the technology as dual-use: it could help find and patch weaknesses, while also potentially assisting governments and cybercriminals with attacks, theft, extortion, or fraud.
- The US has sanctioned Qihoo 360 since 2020 over alleged ties to the Chinese military; Qihoo denies the allegation.
Article Details
- Event Type
- Qihoo 360 announced its Tulongfeng vulnerability-discovery system, which Zhou Hongyi claimed is nearly capable of matching Anthropic’s Mythos. Tulongfeng’s capabilities have not been independently verified.
- Impact
- Zhou Hongyi claimed Tulongfeng had discovered over 3,000 vulnerabilities, many classified as high risk. The article describes advanced AI vulnerability discovery and exploit chaining as potentially strengthening both offensive and defensive cyber capabilities, including against critical infrastructure, while also increasing risks from cybercrime. The US has sanctioned Qihoo 360 since 2020 over alleged ties to the Chinese military; Qihoo denies the allegation.
People
Vendors
AnthropicQihoo 360 says its Tulongfeng system is nearly comparable to Anthropic’s Mythos, a model that dramatically accelerated vulnerability discovery and exploit chaining across IT and OT systems.Qihoo 360Qihoo 360 says its Tulongfeng system is nearly comparable to Anthropic’s Mythos, a model that dramatically accelerated vulnerability discovery and exploit chaining across IT and OT systems.
Products
Claude Mythos Previewexpected that other US-based models would come close to the performance of Anthropic’s initial model, Mythos Preview, within weeks, and the Chinese models would lag several months behind.MythosQihoo 360 says its Tulongfeng system is nearly comparable to Anthropic’s Mythos, a model that dramatically accelerated vulnerability discovery and exploit chaining across IT and OT systems.Mythos 5Anthropic has already surpassed the performance of Mythos Preview, with the release of Mythos 5.TulongfengQihoo 360 says its Tulongfeng system is nearly comparable to Anthropic’s Mythos, a model that dramatically accelerated vulnerability discovery and exploit chaining across IT and OT systems.
Countries
ChinaThe development underscores a global AI-driven arms race in cybersecurity, with China, the US, and criminal actors all seeking advantage from powerful vulnerability-finding capabilities.United StatesQihoo has been sanctioned by the US government since 2020 for ties to the Chinese military, a claim that Qihoo denies.