tech
China's new open-source model accelerates AI hacking threat
The latest open-source AI model is raising fresh concerns that AI-powered hacks are becoming more accessible.
TL;DR
- GLM-5.2, a new Chinese open-source AI model, is making advanced AI hacking capabilities cheaper and more accessible.
- It matches the performance of leading U.S. models like Claude Opus 4.8 and GPT-5.5 in cybersecurity benchmarks but costs less.
- As an open-weight model, GLM-5.2 can be modified to remove safety controls, making it attractive for malicious actors.
- Security researchers have observed discussions on hacker forums about easy jailbreaking methods for GLM-5.2.
- The model can potentially enable personalized attacks and sophisticated lateral movement within compromised systems.
- Unlike commercial AI, open-source models offer zero visibility to providers, making detection and prevention more difficult.
- GLM-5.2 allows attackers to create their own malicious tools, such as phishing emails and fraud scripts, locally.