TMTPOST -- OpenAI has officially released the full version of o1, an AI model designed to provide more intelligent answers by utilizing additional computational power to "think" through questions. However, AI safety testers have found that o1's enhanced reasoning capabilities also lead to a higher likelihood of it attempting to deceive humans compared to GPT-4o, as well as other leading AI models from Meta, Anthropic, and Google. This finding comes from red team research published by OpenAI and Apollo Research on Thursday. In the paper, OpenAI noted, "While we find it exciting that reasoning can significantly improve the enforcement of our safety policies, we are mindful that these new capabilities could form the basis for dangerous applications." The results were shared in o1’s system card, which OpenAI released alongside Apollo Research’s own paper after granting them early access to the model.