The release of GLM-5.2, an open-source model developed by Z.ai, has raised concerns among security researchers regarding the accessibility of advanced AI hacking capabilities. Released on June 18, 2026, GLM-5.2 is noted for its capabilities that compete with models such as Claude Opus 4.8 and OpenAI's GPT-5.5, while being less expensive to operate.
Security evaluations from Graphistry and Semgrep indicate that GLM-5.2 performs comparably to leading U.S. models in cybersecurity investigations and vulnerability discovery. There are allegations from Graphistry that GLM-5.2 may be an illegal distillation of GPT-5.5 and Opus 4.8, which could explain the rapid advancements of Chinese models in comparison to U.S. counterparts.
Unlike proprietary models, GLM-5.2 allows users to download and modify it, potentially removing safety controls and enabling tailored operations without reliance on commercial providers. Security consultant Jason Baker noted that discussions in Russian-language forums indicate hackers are already exploring ways to utilize GLM-5.2 for hacking tasks.
Travis Lanham, CTO of Armadin, explained that the model enables attackers to personalize their methods once they infiltrate a system, mimicking sophisticated human attacks. Some hackers have reportedly discovered methods to bypass the model's limitations, with simple jailbreaks proving effective.
The open-source nature of GLM-5.2 presents fewer barriers for malicious use compared to commercial models, which typically have mechanisms to detect and ban misuse. Analysts like Roye Bass from Halcyon highlighted that attackers can leverage GLM-5.2 to create phishing emails and other malicious content. However, Baker cautioned that the current AI-generated exploits are not yet highly effective, as the skills required to utilize these technologies have not fully developed.
Z.ai's founder Jie Tang has indicated plans to release an open-source model that could rival Anthropic's Fable by the end of the year, while another Chinese firm, 360 Technology, has also announced the development of tools to match Anthropic's Mythos.