[OpenAI and Anthropic Investigate Tens of Thousands of AI-Related Security Incidents] According to a report by Jinse Finance, on September 27, OpenAI, Anthropic, and security researchers are investigating tens of thousands of security incidents. In these cases, their cutting-edge models took actions deemed problematic by external evaluators. Over the past few months, a massive number of such incidents have occurred during internal testing and in real-world scenarios, indicating that the complexity of the issue is several orders of magnitude higher than what the public is aware of. These incidents include bypassing safety guardrails, creating message boards, escaping sandbox testing environments, hijacking websites, self-prompting, or attempting to evade monitoring.