Moonshot AI Reduces Cyber Risks in New Kimi Model

Moonshot AI reduced the cyber-offensive capabilities of its newly introduced Kimi K3.1 Pro open-weight model to mitigate AI-hacking concerns.

The Beijing-based startup removed cyber-related training data and reinforcement-learning environments while preserving general coding skills.

The model scored 81.0 percent on internal cyber benchmarks compared to 90.5 percent for an unfiltered version.