Moonshot AI says it filtered cyber related training data from its newest open weight model, trimming its hacking ability before weights go public on November 4.
Tamrah News Desk | October 24, 2026
Moonshot AI has reduced the offensive cyber capabilities of Kimi K3.1 Pro, the Beijing based startup said on Saturday, October 24, as concern grows that autonomous AI could make hacking easier. According to Reuters, the company removed cyber related training data before launch and plans to publish the model’s open weights by November 4.
What Moonshot Changed in Kimi K3.1 Pro
Reuters reported that Moonshot AI removed cyber related data from the training corpus and the reinforcement learning environments used to build the model, while keeping general coding skills. The company calls Kimi K3.1 Pro its most capable open weight model in the 3 trillion parameter class. Weights are the files that let outside developers run and modify a model on their own systems.
Reuters quoted the company as saying it aims not to “extend the absolute frontier of cyber capability” through open weights. Moonshot’s launch blog says cyber related training data was removed from both training stages. It adds that the company intends to raise the cyber capabilities available in open weights step by step.
How Much Hacking Ability Did Kimi K3.1 Pro Lose?
Moonshot said Kimi K3.1 Pro scored 81.0 percent on its internal cyber benchmark, against 90.5 percent for a similar version built without the risk reduction steps, Reuters reported. The test involves finding and exploiting software vulnerabilities in a controlled environment. Moonshot said the gap shows that filtering training data reduced the model’s offensive performance, although the model still beat its predecessor.
The blog states that the model’s cyber capabilities remain behind those of leading proprietary models that have been publicly available for at least three months. Reuters noted that the results have not been independently verified.
Why Open Weights Raise the Stakes for AI Cyberattacks
Reuters reported that governments, technology companies and security researchers worry that AI able to write code, operate software tools and finish long tasks could help users find vulnerabilities and automate parts of attacks. Once open weights are widely distributed, Reuters said, the original developer has less ability to restrict access, monitor use or update safeguards than with a hosted service.
Moonshot argued that open weight models can also improve security. According to Reuters, the company said they help more organisations inspect and secure their own software, including older systems that receive little attention.
China Export Controls Add Policy Pressure
Reuters reported that 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. TechNode reported that the draft was issued on October 16 by the Ministry of Commerce and the Ministry of Science and Technology. It covers 23 AI technologies and is open for public comment until November 15.
TechNode added that technology already public through open source releases or academic papers would not be covered by the export controls.
Kimi K3.1 Pro Performance and Availability
The Decoder reported that Kimi K3.1 Pro scored 55 points on the Artificial Analysis Intelligence Index, placing fourth just behind Claude Opus 5.5. It put the model at 2.8 trillion parameters and said API prices fell by 33 percent.
Moonshot said Kimi K3.1 Pro and a faster Highspeed version, which runs two to three times quicker, are available on Kimi.ai, Kimi Work, Kimi Code and the Kimi API. Reuters said the model has a context window of one million tokens.
Wider Debate Over Open Weight Cyber Capabilities
In September, the US National Institute of Standards and Technology said Z.ai’s GLM 5.3, released on August 14, was the most cyber capable open weight model released to date. It also found the model trailed leading US systems by about four months. NIST said Kimi K3 had been the previous best Chinese model on most of its evaluations.
Technology.org reported that on October 6 Anthropic merged Project Glasswing into a three tier Cyber Verification Program for vetted defenders.
What Happens Next for Kimi K3.1 Pro?
Moonshot says the full weights will arrive by November 4. The public comment period on China’s draft export controls closes on November 15.
Conclusion
The Kimi K3.1 Pro launch shows how Moonshot AI is trying to balance a powerful open weight model against growing fears of AI cyberattacks. The company says filtered training data lowered the model’s hacking ability, although its internal benchmark results are not independently verified. Readers following the Kimi K3.1 Pro open weights release date, China export controls and open weight cyber capabilities will find confirmed updates here as they emerge.
