News
Moonshot AI Reduces Cyber-Offensive Capabilities in Kimi K3.1 Pro Model
Beijing-based startup Moonshot AI announced it intentionally reduced the cyber-offensive capabilities of its Kimi K3.1 Pro AI model. The company stripped cyber-related data from the model's training corpus to limit its ability to exploit software vulnerabilities while maintaining coding skills. On Moonshot's internal exploit benchmark, Kimi K3.1 Pro scored 81.0%, down from 90.5% for an unfiltered version. The model architecture uses 2.8 trillion total parameters with 104 billion active parameters and a 1-million-token context window. Moonshot plans to publicly release the model's full weights by November 4, 2026.