Moonshot announces changes to its new Kimi AI model intended to reduce hacking-related capabilities as concerns grow about AI-enabled cyberattacks. Across the reports, the focus is on curbs to the model’s access to or exposure to cyber content used during development.
According to Investing.com, the company’s updates are presented in the context of broader fears over AI-powered cyberattacks, while indicating that Moonshot has tightened how the model is trained and tested. Rappler adds that Moonshot removes cyber-related data not only from the training corpus but also from reinforcement-learning environments used to develop the model, while aiming to keep general coding skills. Free Malaysia Today similarly reports the removal of cyber-related data from training materials and testing environments.
Taken together, the outlets converge on the core claim: Moonshot modifies training and evaluation inputs to limit cyber capability, while retaining other general programming abilities.
