The gist: Open-source AI model with local deployability and halved operational costs reduces access barriers to cyberattack tools for a broader spectrum of actors.
Chinese company z.AI has released the open-source model GLM-5.2, which matches US systems on security tasks such as vulnerability detection and can be operated locally without cloud surveillance. This significantly lowers the technical and economic barriers to automated cyberattacks.
According to analyses by Graphistry and Semgrep, GLM-5.2 achieves performance on par with leading models from OpenAI, Google and Anthropic on cybersecurity tasks – at roughly half the operational costs according to manufacturer specifications. The model is suitable for software flaw detection, security analysis and similar specialized security tasks.
The crucial difference from commercial US models: GLM-5.2 runs on local computers instead of cloud infrastructure. This eliminates any external monitoring by cloud providers and their abuse prevention mechanisms. According to media reports, hackers are already discussing methods to completely remove remaining security barriers. Security researchers also demonstrate that the model can be prompted through simple input to ignore internal safeguards.
This availability enables attackers to systematically experiment with AI-driven methods – such as writing phishing emails or automated searches for software vulnerabilities. Capabilities that were previously reserved for specialized actors are now becoming more widely accessible.
Security experts like Gene Moody of Action1 view this less as a technical breakthrough than as a geopolitical escalation: China’s move marks another arms race in the digital realm, in which attackers and defenders alike gain access to increasingly powerful tools. The strategic problem is that the technology is being developed not as a common defensive field, but as a competitive weapon.
Source: www.it-daily.net · Published 2 July 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification via Lumi News Pipeline v1.7.2.