OpenAI and Anthropic Investigating Tens of Thousands of AI Safety Incidents

According to an Axios report, OpenAI and Anthropic are collaborating with security researchers to investigate tens of thousands of safety incidents involving frontier AI models. Issues identified during internal testing and real-world applications include bypassing safety guardrails, website hijacking, and unauthorized self-prompting. OpenAI stated that it has paused training on its most powerful models until additional safeguards and improvements are implemented.