Chinese AI matches Anthropic’s Mythos in cybersecurity

Chinese artificial intelligence systems have achieved performance comparable to Anthropic's Mythos model in cybersecurity vulnerability discovery, according to a Wall Street Journal News Corp ↘ 1.33% report published June 27, a development that threatens to undermine Washington's strategy of restricting access to frontier AI capabilities.threads+1

Zhipu AI's GLM-5.2 Closes the Gap

Zhipu AI, the Beijing-based company rooted in Tsinghua University, released its GLM-5.2 model — a 744-billion-parameter open-weight system — which independent testing shows can match Anthropic's proprietary Mythos on specific cybersecurity benchmarks. Security research firm Graphistry found that GLM-5.2 achieved a 28/59 solve rate on the CyBT-CTF security agent benchmark, tying proprietary models from Anthropic and OpenAI while operating at roughly half the cost. Research firm Artificial Analysis rated it the third most capable AI overall, behind Anthropic's Claude Fable and OpenAI's ChatGPT 5.5, but ahead of Google Alphabet Inc.'s ↗ 1.53% Gemini.mlq+3

The model ships under a permissive MIT license with a one-million-token context window, meaning it can process entire codebases at once — a capability with direct implications for automated vulnerability hunting.reddit+1

360 Security Unveils "China's Version of Mythos"

At the ISC.AI 2026 conference in Beijing on June 24, 360 Security Technology founder Zhou Hongyi introduced two AI-powered cybersecurity tools: Tulongfeng for automated vulnerability discovery and Yitianzhen for cyber defense and incident response. Zhou explicitly called Tulongfeng "China's version of Mythos," claiming the system has identified 3,432 software flaws, with 105 confirmed by Chinese government authorities. Zhou acknowledged a 20-30% capability gap compared to leading US models but argued that building a professional attack-and-defense team mattered more than relying on a single advanced AI.scworld+1

Alibaba Accused of Infiltrating Claude

The developments coincide with Anthropic's accusation, detailed in a letter reviewed by the Wall Street Journal, that Alibaba and its Qwen AI division ran what Anthropic called "the most significant known distillation attack" against Claude. According to the letter sent to US senators and White House officials, Alibaba-linked operators created nearly 25,000 fake accounts and conducted 28.8 million conversations with Claude between April 22 and June 5, 2026. The Financial Times and Bloomberg independently confirmed the allegations.citybiz+4

The multi-pronged challenge — domestic model development, open-source distribution, and alleged illicit access to US systems — arrives as US export controls have forced Anthropic to suspend international access to its frontier models. As Axios reported, advanced AI hacking capabilities are becoming "dramatically cheaper and more accessible" through open-source Chinese models, raising questions about whether Washington's technology containment strategy can keep pace.fortune+1