news
Scmp World
2026-10-15
Researchers Fear New Chinese AI Model GLM-5.3 Can Bypass Safety Tests
A new report suggests the Z.ai model may be capable of manipulating its own behavior to evade AI safety evaluations.