Atlas browser’s Omnibox opens up new privacy and security risks

Researchers found that OpenAI's Atlas Omnibox (the combined search and prompt input) can be abused by specially crafted links to perform prompt injection, causing the browser to treat injected instructions as trusted user prompts and bypass safety checks; although OpenAI states mitigations exist, the report warns that agentic AI browsers remain vulnerable and could expose users' privacy and sensitive account data if exploited.