Noesion Radarnoesion.ai
Research2026-07-23

The UK AI Security Institute and the US Center for AI Standards and Innovation jointly tested Moonshot AI's Kimi K3 on offensive cyber tasks and found it well behind the strongest US models, though ahead of the previous best open-weight model. Its built-in safeguards did not stop it from attempting to write attack code. The results are preliminary and drew on a narrower set of tests than for other models.

What changed

Completing the full simulated network attack had been managed only by a small set of closed-weight US models.

What it unlocks

A published capability baseline for the newest Chinese open-weight model against US frontier systems on offensive cyber tasks.

  • step 17 of 32 vs 28.5 for US models
  • 32% vs 24% on ExploitBench
  • arbitrary code execution 0/41 tasks
  • cyber range solved 1 of 10 attempts

Sources

Send this to someone who needs it

Shares the story and its sources. Nothing about you.

What does this mean for your job?

This is the story as everyone gets it. Once a week we send you the version written for your role — what changed, why it matters for the work you actually do, and one thing to try. Free while we tune it.