AI-ul chinezesc care a stârnit îngrijorare: cercetătorii spun că poate fi manipulat pentru arme biologice și asasinate

The Kimi-K3 model by Moonshot AI has been found to have significant vulnerabilities, potentially allowing it to provide dangerous information. Researcher Peter Garrigan revealed that the model could be manipulated to offer instructions on creating biological weapons and planning assassinations. He noted that similar vulnerabilities exist in AI models developed in the U.S. as well. Moonshot AI is conducting an internal investigation into these claims, but has not confirmed the findings publicly. The situation raises concerns about the broader implications of AI systems becoming more capable and accessible.