A new Chinese open-source AI model, GLM-5.2, is raising alarms that advanced hacking capabilities are becoming cheaper and more accessible. Released last week by Z.ai, the model reportedly matches the performance of Claude Opus 4.8 and GPT-5.5 on cybersecurity benchmarks, while costing roughly half as much to run.
Unlike proprietary systems such as Claude or ChatGPT, open-weight models like GLM-5.2 can be downloaded and modified directly. This allows users to strip away safety controls, making it easier to automate and personalize malicious attacks against networks and systems.
Two independent security evaluations — from Graphistry and Semgrep — found GLM-5.2 performing on par with leading U.S. models on vulnerability-discovery tasks. Graphistry also suggested the model may be an “illegal distillation” of both GPT-5.5 and Opus 4.8, which could explain how Chinese AI has rapidly closed the gap.
The implications are stark: as the barrier to entry drops, more actors — from cybercriminals to state-sponsored groups — could gain access to sophisticated AI-driven attack tools. Z.ai did not respond to a request for comment on the distillation allegation or safety controls.
The claims of intellectual property theft remain unproven, and Z.ai has not confirmed or denied them. Some analysts caution that performance benchmarks may not fully translate into real-world autonomous hacking scenarios, and that existing defenses could still mitigate many AI-enhanced threats.