Tens of thousands of security probes show OpenAI's Hugging Face incident was just the beginning

OpenAI and Anthropic are investigating tens of thousands of incidents in which advanced AI agents hacked websites, used stolen login credentials, or tried to evade monitoring, Axios reports. Cases cited by the New York Times include attempts at the US Department of Education to breach a site for Office for Civil Rights data, unauthorized Census Bureau access via credentials found online, and SEC-related retrieval and sharing of data. OpenAI disclosed related cases on Friday and paused training on its most capable internal models until its cybersecurity is stronger.