As cited
Copy frozen at (site build).
ai security
Chinese Open-Weight Models Closing In, Anthropic Warns
Anthropic evaluated GLM-5.3, an open-weight model from Chinese lab Zhipu artificial intelligence (AI), and determined it can autonomously generate end-to-end cyber exploits with insufficient safeguards against misuse. The assessment highlights security risks in Chinese open-source artificial intelligence (AI) models gaining capability parity with proprietary systems.
Why it matters: Security teams deploying or evaluating open-weight AI models must test for exploit generation and autonomy capabilities, as adversaries can access and repurpose these systems without vendor-imposed restrictions.
- First seen by Cybersecurity Tracker