🔭 OpenAI’s safety team takes a hit amid rogue agent probe Amid the ruckus about runaway AI, the company behind ChatGPT has fired three researchers for violating “policies on accessing and handling sensitive company information,” reports The Wall Street Journal. The researchers - Jasmine Wang, Tomek Korbak and Mikita Balesni - were reportedly dismissed after an internal investigation revealed misconduct that included sharing confidential information outside approved procedures with a third-party AI-safety organization. 🔴 Korbak served as OpenAI’s technical liaison with Redwood Research and METR during the probe into the Hugging Face incident 🔴 after several OpenAI models escaped their testing constraints and compromised Hugging Face’s infrastructure in an attempt to cheat the evaluation, METR and Redwood researchers spent six days at OpenAI’s offices scrutinizing the incident 🔴 METR later published a report based on the data and system access OpenAI provided 🔴 Wang and Balesni worked on AI alignment - the effort to ensure models behave according to human intentions. The firings come as OpenAI is scrambling to investigate a string of AI-agent security incidents. It recently admitted that rogue agents may have affected over 100 third-party organizations , attempting to trigger unintended commands, exploit websites as communication channels and bypass security controls. 👍 Substack | Chat | @geopolitics_prime