Chinese AI startup Moonshot AI has reduced the offensive cyber capabilities of its latest artificial intelligence model, Kimi K3.1 Pro, amid growing concerns that increasingly autonomous AI systems could make sophisticated cyberattacks easier and cheaper to execute.The Beijing-based company announced the safeguards on October 24, describing Kimi K3.1 Pro as its most capable open-weight model in the three-trillion-parameter class. Moonshot plans to release the model’s weights, which allow developers to run and modify it on their own systems, by November 4.Kimi K3.1 Pro Cybersecurity SafeguardsMoonshot said it removed cyber-related data from the model’s training corpus and reinforcement-learning environments to limit potential misuse. The company said it had retained the model’s general coding capabilities while reducing its ability to carry out offensive cyber operations.“We aim not to release open weights that would extend the absolute frontier of cyber capability,” the company said.The decision reflects growing concerns among governments, technology companies and security researchers that advanced AI could help users identify software vulnerabilities, write malicious code and automate complex attacks involving multiple steps and software tools.Benchmark Shows Reduced Offensive CapabilitiesMoonshot reported that Kimi K3.1 Pro scored 81% on its internal cyber benchmark, compared with 90.5% for a similar version developed without the risk-reduction measures.The test involved identifying and exploiting software vulnerabilities in a controlled environment. According to Moonshot, filtering cyber-related data reduced the model’s offensive performance, although it continued to outperform its predecessor.The results have not been independently verified.Open-Weight AI Creates Security ChallengesMoonshot argued that open-weight AI models can also strengthen cybersecurity by allowing organisations to examine and secure their own software, including older systems that may receive limited maintenance.However, releasing model weights creates challenges for oversight. Once widely distributed, the files are harder for the original developer to monitor or restrict. Developers also have less ability to update safeguards or control access than they do with AI systems hosted through proprietary services.China Considers Controls on AI Cyber ToolsThe announcement comes as China considers proposed export controls targeting AI-driven malware-generation and mutation technology, alongside restrictions on AI-assisted vulnerability discovery.The draft measures, which remain open for public comment, cite concerns that these capabilities could be used against critical infrastructure, defence-industrial systems and core government networks, or to bypass security protections.Kimi K3.1 Pro also features a one-million-token context window, enabling it to process large volumes of information in a single context. Moonshot has separately released K3.1 Pro Highspeed, which it says operates two to three times faster than the standard version.The model’s release highlights the challenge facing AI developers: preserving the benefits of powerful coding systems while limiting their potential to enable cyberattacks.
Click Here For The Original Source.
——————————————————–
