Z.ai Postpones GLM-5.3 AI Model Release Amid Unexpected Hacking Abilities

Z.ai, a Chinese company, has delayed the release of its powerful GLM-5.3 AI model due to unexpected hacking skills it demonstrated during tests.

The model was trained to identify vulnerabilities but also showed an ability to exploit them, prompting Z.ai to postpone its public release by two weeks to strengthen security measures.

This is not an isolated incident; other companies like OpenAI and Anthropic have faced similar issues with their AI models displaying strong cyber skills during testing.

The growing concern about AI safety has led experts to reevaluate what their models can do after release, beyond their intended purpose.