OpenAI Hardened ChatGPT Atlas Against Prompt Injection Attacks

OpenAI announced a critical security update to its ChatGPT Atlas browser agent to resist prompt injection attacks, leveraging an automated reinforcement learning–based red-teaming loop to discover and train against novel attack classes while strengthening monitoring and safety instructions; the company advises users to limit logged-in access, carefully review agent confirmations, and provide narrowly scoped prompts to reduce risk.