Beijing-based Moonshot AI said on Saturday that it had weakened the offensive hacking abilities of its newest model. The announcement came as concern rises that increasingly autonomous artificial intelligence could sharply speed up cyberattacks.
The company unveiled the curbs alongside Kimi K3.1 Pro, which it describes as its most capable open-weight model in the 3-trillion-parameter class. Moonshot plans to publish the model’s weights, the files that let outside developers run and adapt it on their own systems, by November 4.
How Moonshot stripped cyber material from Kimi K3.1 Pro
Moonshot said it removed cyber-related data from both the training corpus and the reinforcement-learning environments used to build the model. It said general coding skills were kept. “We aim not to release open weights that would extend the absolute frontier of cyber capability,” the company said.
Governments, technology companies and security researchers are increasingly worried that AI could make sophisticated hacking easier and cheaper. Models that write code, operate software tools and handle long, multi-step tasks may help users find weaknesses in programs and automate parts of an attack.
China’s draft export controls on AI hacking tools
Last week China proposed export controls that would ban exports of AI-driven malware generation and mutation technology. The draft would also restrict AI-assisted tools for discovering software vulnerabilities. The measures remain open for public comment.
The draft cites the risk that such capabilities could be aimed at critical infrastructure, defence-industrial systems and core government networks, or used to bypass security protections. Moonshot argued that open-weight models can improve security by letting more organisations inspect and protect their own software, including older systems that receive little attention. Still, open releases make central oversight harder than closed, proprietary systems do. Once files spread widely, the original developer has less ability to restrict access, monitor use or update safeguards than it would with a hosted service.
Benchmark results for Kimi K3.1 Pro
The company said K3.1 Pro scored 81.0 per cent on its internal cyber benchmark. A similar version without the risk-reduction measures scored 90.5 per cent. The test involved finding and exploiting software vulnerabilities in a controlled environment.
Moonshot said the gap showed that filtering cyber-related data lowered the model’s offensive performance, although it still outperformed its predecessor. The results have not been independently verified. K3.1 Pro has a 1-million-token context window, allowing it to process very large volumes of text at once. Moonshot also released K3.1 Pro Highspeed, which it said runs two to three times faster than the standard model.
