Chinese AI startup Moonshot AI is deliberately limiting the hacking capabilities of its latest artificial intelligence model, highlighting a growing dilemma for the industry: how do companies make AI more powerful without making cyberattacks easier to carry out?
The company introduced Kimi K3.1 Pro on October 24, describing it as a highly capable, open-weight model with around 3 trillion parameters. Unlike conventional AI services that operate entirely through a provider’s servers, open-weight models allow developers to download the underlying model files, run them on their own systems, and modify how they work.
Moonshot said it removed cyber-related data from the model’s training dataset and reinforcement-learning environments to reduce its ability to conduct offensive cyber operations, while retaining its general coding capabilities. The company plans to release the model’s weights on November 4, opening it up to a wider developer community.
Moonshot reduces hacking capabilities without sacrificing coding
Moonshot’s internal testing suggests the safeguards have made a measurable difference. Kimi K3.1 Pro scored 81% on the company’s cybersecurity benchmark, which involved finding and exploiting software vulnerabilities in a controlled environment. A similar version without the risk-reduction measures scored 90.5%.
That represents a 9.5-percentage-point drop in the reported score. However, the results have not been independently verified, and the model reportedly still outperformed its predecessor.
The distinction matters because the same AI capabilities can be used for both defence and offence. Security teams could use AI to identify weaknesses in software before criminals exploit them. Attackers, meanwhile, could use models that write code, operate software tools and complete multi-step tasks to automate parts of an attack.
Moonshot argues that open-weight AI can help organisations inspect and secure neglected or older software. But once model files are distributed widely, the original developer has less control over who uses them and whether safeguards remain effective.
AI security concerns are becoming a global policy issue
Moonshot’s decision comes as governments and technology companies worry that increasingly capable AI could make sophisticated cyberattacks cheaper and easier to execute. China has also proposed export controls targeting AI-driven malware-generation technology and certain AI-assisted vulnerability-discovery tools. The proposals remain subject to public comment.
For businesses and ordinary users, the stakes extend beyond individual computers. More automated attacks could increase pressure on organisations responsible for protecting personal information, financial systems and critical infrastructure. At the same time, restricting AI security research too aggressively could leave vulnerabilities undiscovered.
Kimi K3.1 Pro also features a one-million-token context window, allowing it to process large amounts of information in one go. Moonshot has introduced a Highspeed version that it says runs two to three times faster than the standard model.
The next major test will come when Moonshot releases the model weights on November 4. Developers and security researchers will then have greater access to examine its capabilities, while the industry will face a difficult question: can safeguards remain effective once a powerful AI model is available for independent use?
RELATED COVERAGE:
