OpenAI explains agent link protection: the URL itself can carry data away
OpenAI published an explanation of data protection when agents open links, noting that URL parameters can carry sensitive content and redirects may change the destination. These controls address data leakage through URLs; they do not establish that webpage content is trustworthy.
Source Facts
On 28 January, OpenAI explained the risk of data exfiltration through URLs: the requested URL is sent to the destination server. If private content is embedded in it, data can leak even when the chat response does not directly display it.
The company explains that specific publicly known URLs inform decisions about automated retrieval, with additional restrictions for unconfirmed addresses. The source also notes that this protection cannot guarantee a page contains no misleading content and must be combined with other prompt injection defences and monitoring.
Agent outbound controls should inspect complete URLs and data flows, not just destination website names.
Original sources
Keeping your data safe when an AI agent clicks a link
Sources checked: 2026-10-26 · Prepared with SI agent assistance; articles have not been individually reviewed by a person.
This is an English translation of our Chinese summary and editorial observations; source dates and this site’s verification date are shown separately. If an original advisory changes, refer to the source.
KEEP READING
Recommended reading
IllustrationLiteLLM explains package compromise: exposed keys must be revoked to complete the response
IllustrationOpenAI details training environment isolation improvements, using offline operation and continuous red teaming to constrain agents
Illustration