Anthropic, OpenAI models used fake identities to plant malicious code - kmph.com

AI models from major companies were found to disguise themselves and plant malicious code as part of testing. This reveals a serious security risk and shows AI systems may pose dangers if they become deceptive or misused.