Zhipu AI's GLM-5.2 matches Mythos as firm raises $4 billion

Beijing-based Zhipu AI released an open-weight model matching Anthropic's Mythos 5 on cybersecurity benchmarks while raising $4 billion in a Hong Kong share placement.

Zhipu AI's GLM-5.2 open-weight model matches Anthropic's Mythos 5 on cybersecurity benchmarks, giving hackers and researchers free access to frontier AI capabilities that the US government briefly deemed a national security risk.

"An attacker can run it locally without safety guardrails, fine-tune it against their specific targets, and operate with zero visibility to any provider or defender," Travis Lanham, chief technology officer and founder of AI cybersecurity firm Armadin, said.

The model, released by Beijing-based Z.ai, scored competitively against Anthropic's Mythos 5 in bug detection and code vulnerability identification, according to benchmarks from security firms Semgrep and Graphistry. Semgrep titled its analysis "We Have Mythos at Home." Unlike Mythos 5, which the Trump administration forced Anthropic to withdraw in June before approving a limited re-release to about 100 US organizations, GLM-5.2 can be downloaded by anyone and run on virtually any hardware with no vendor oversight. Researchers at Graphistry suggested Z.ai may have distilled OpenAI's GPT-5.5 and Anthropic's Opus 4.8, a practice that could explain how the Chinese lab caught up so quickly to models that cost hundreds of millions of dollars to train.

The dual development positions Zhipu AI as a serious contender in the global AI arms race. The company's stock has surged about 1,500 percent since its Hong Kong IPO in January, and the placement at HK$1,588 to HK$1,698 per share represents a 7 percent to 13 percent discount to the July 8 close. Proceeds will fund research and development, global expansion, and mergers and acquisitions, according to the term sheet seen by Reuters.

Open-Source AI Poses New Cybersecurity Challenge

Consultants told Axios that hackers are already trading GLM-5.2 jailbreaks on Russian-language forums, suggesting the threat is immediate. The model's open-weight architecture means bad actors can strip safety guardrails, fine-tune it for malicious code generation, and operate without detection — a scenario that US regulators had sought to prevent by restricting Mythos 5's availability to about 100 vetted organizations.

The $4 billion raise is part of a broader wave of Chinese technology firms tapping Hong Kong's capital markets. CATL raised about $5 billion in a similar placement in April, while Biren Technology completed a $900 million share sale and Muxi is seeking about $850 million. Zhipu AI's offering, led by China International Capital Corp. as sole coordinator, shows investor appetite for Chinese AI equities despite geopolitical tensions over technology transfer and national security.

For investors, the emergence of GLM-5.2 compresses the timeline for AI commoditization. If open-weight models can match frontier performance without the safety infrastructure of companies like Anthropic, the competitive moat for proprietary AI narrows — and the cybersecurity risk premium for enterprises relying on these models widens. Zhipu AI's valuation, implied by its 1,500 percent rally since its January IPO, already prices in aggressive growth expectations. Whether the market has fully priced in the regulatory and security blowback from open-weight distribution remains an open question, particularly as US-China technology tensions continue to escalate.

This article is for informational purposes only and does not constitute investment advice.