Moonshot Kimi K3.1 Pro Curbs Hacking Before Open Release
Moonshot stripped hacking skills from Kimi K3.1 Pro before its open-weight release on November 4. Here is why the Chinese AI lab took that safety step.
Moonshot AI has cut back the hacking ability of its newest model, Kimi K3.1 Pro, before releasing it to the public. The Beijing-based startup announced the move on October 24, as it introduced the model it calls its most capable open-weight system yet. The company plans to publish the model's weights, the files that let outside developers run and change it on their own systems, by November 4. Before that happens, Moonshot deliberately removed cyber-related data from the model's training.
What Moonshot Announced
Kimi K3.1 Pro is described by Moonshot as its most capable 3-trillion-parameter-class open-weight model. In plain terms, it is a very large AI system that developers will soon be able to download and run themselves, instead of only using it through Moonshot's own service. The model is already available through Moonshot's apps and API, with a faster companion version called Kimi K3.1 Pro Highspeed. The full open weights follow by November 4. Moonshot said it removed cyber-related data from the training corpus and from the reinforcement-learning environments used to develop the model. At the same time, it kept the model's general coding skills intact. "We aim not to release open weights that would extend the absolute frontier of cyber capability," the company said.
Why the Curbs Matter Now
AI models that can write code, operate software tools and complete long, multi-step tasks could help users find weaknesses in programs and automate parts of cyberattacks. Governments, tech companies and security researchers are growing more worried that increasingly capable AI could make sophisticated hacking easier and cheaper. That worry is now shaping policy too. China proposed export controls last week that would ban exports of AI-driven malware-generation and mutation technology and restrict AI-assisted vulnerability-discovery tools. The draft measures remain open for public comment, and they cite risks to critical infrastructure, defence-industrial and core government systems, and to security protections. Moonshot's move lands in the middle of that debate. An open-weight release is different from a hosted service: once the files are widely shared, the original developer has less ability to restrict access, monitor use or update safeguards.
What K3.1 Pro Can Still Do
The curbs do not make the model weak. Moonshot's own benchmark numbers put K3.1 Pro close to the best closed models on coding work. On Terminal-Bench 4.0, a test that scores coding agents on realistic terminal tasks, Moonshot reports 65.7 percent for K3.1 Pro. That is ahead of OpenAI's GPT-6 Astra at 58.2 percent and just behind Anthropic's Claude Opus 5.5 at 66.4 percent. These are Moonshot's own figures, not yet verified by an independent lab. The model itself is built as a mixture-of-experts design with roughly 2.8 trillion total parameters and 104 billion active at any one time, and it handles a 1-million-token context window. Moonshot also argues there is a positive side to open weights. Open models can help more organisations inspect and secure their own software, including older systems that receive little attention.
What Happens Next
The key date is November 4, when Moonshot has promised to publish the full weights. Until then, the main test is whether independent researchers can reproduce Moonshot's benchmark claims and check how far the hacking curbs really go. If the weights ship as promised, K3.1 Pro would become one of the strongest open-weight models available, and a test case for how AI labs handle the safety trade-off of open releases. If other labs follow Moonshot's lead, removing the most dangerous cyber skills before an open release could become the new normal.
Sources
- Reuters - https://www.reuters.com/technology/moonshot-curbs-hacking-ability-new-kimi-model-fears-grow-over-ai-powered-2026-10-24/ - SiliconANGLE - https://siliconangle.com/2026/10/23/moonshot-ai-launches-new-flagship-kim-k3-1-pro-model-with-3t-parameters/