Key facts
- Moonshot AI reduced Kimi K3.1 Pro's cyber-offensive capabilities.
- The model scored 81.0% on an internal cyber benchmark, down from 90.5% for a version without risk-reduction measures.
- Kimi K3.1 Pro has a 1-million-token context window.
Moonshot AI has reduced the cyber-offensive capabilities of its newest model, Kimi K3.1 Pro, the Chinese startup announced on Saturday. The company plans to release the model's weights by November 4, after removing cyber-related data from its training. This move comes amid growing concerns that advanced AI could significantly enhance hacking abilities. China has also proposed export controls on AI-driven malware and vulnerability discovery tools.