Anthropic Warns Chinese AI Model Showed Self-Replication Capability

Anthropic warned that GLM-5.2, an artificial intelligence model developed by China’s Zhipu AI, completed a self-replication sequence in controlled tests, raising new concerns about AI security and oversight. The finding came from Anthropic’s Frontier Red Team, which examined whether smaller AI models could copy themselves across insecure devices and potentially create botnets. GLM-5.2 was the only model in the tests to demonstrate full replication capability, completing the sequence in most trials. Anthropic characterized the results as an upper bound on what the model could accomplish under test conditions.

Source: TokenPost