China’s New Zhipu AI Matches Claude Mythos’ Security Detection, Is Anthropic Screwed?

It’s OK, don’t freak out just yet, Anthropic CEO Dario Amodei

Claude and ChatGPT are still better all-purpose AI systems, BUT Zhipu’s latest model GLM-5.2 is on a par with (if not better than) Mythos “in specific cybersecurity and software vulnerability detection tasks.”

It’s a worrying thing to hear, particularly if you’re the American government that wants to keep the cutting edge of AI research red, white, and blue and screeching like an eagle.

Because GLM-5.2 is available worldwide under an ‘open-weight license,’ meaning that any developer can download and run the model on off-the-shelf tech. Mythos, on the other hand, is limited by America’s current export rules.

Z.ai Vs Anthropic, the biggest fight since Jake Paul killed Mike Tyson

The face-off came about when Semgrep tested both models and scored GLM-5.2 a vulnerability detection score of 39% but Claude only got a 32-37%. Embarrassing.

But more importantly than accuracy is price. Zhipu’s model came in at $0.17 per vulnerability, a sixth of Claude’s cost.

Here’s the full comparison breakdown (credit: Cyber Security News):

Metric	GLM-5.2 (Zhipu AI)	Claude Mythos (Anthropic)
IDOR Detection F1 Score	39%	~32–37%
Cost Per Vulnerability Found	~$0.17	~$1.00+
Access Model	Open-weight (public)	Restricted / export-controlled
General-Purpose Benchmark Rank	Trails U.S. models	Frontier-tier
License	Permissive	Proprietary

What was once a monopoly in the US of these powerful AI tools now looks like a more even playing field.

You see, Claude’s security tools are limited in China since the White House sees them as a potential security risk. But now with a newer, better security AI, China doesn’t need your sinking Claude anyway.

What even is a Zhipu?

In case you’ve never heard of Zhipu AI before, it’s very simple: the company actually rebranded last year to become simply ‘Z.ai’ (not to be confused with Elon’s xAI) but officially it’s called Knowledge Atlas Technology Joint Stock Co., Ltd., or 北京智谱华章科技有限公司 in Chinese which actually translates to Beijing Zhipu Huazhang Technology Co., Ltd.

Got all that?

Founded in 2019, it’s now China’s third-largest LLM company despite being blacklisted in the US.

But now things might be changing, as this new security development could easily mean that people around the world will soon know the name ‘Zhipu AI (Z.ai) Knowledge Atlas Technology Joint Stock Co., Ltd. 北京智谱华章科技有限公司 (Beijing Zhipu Huazhang Technology Co., Ltd.)’.