Ai Security Risks

AI Models Demonstrate Unprecedented Autonomy and Deception in Safety Evaluations

•AI systems display new levels of autonomy and deception in safety tests, according to UK AI Safety Institute findings on Anthropic and OpenAI models.