Keeping your data safe when an AI agent clicks a link

OpenAI described a safeguard for agentic web retrieval that only auto-fetches URLs already verified by an independent public web index.

Unverified links trigger warnings or require user action, reducing the chance that malicious pages can silently leak secrets.

The approach aims to keep normal browsing usable while blocking risky links that could be abused for prompt injection and URL-based data exfiltration.