Latest
Updated hourly · UK time
ilwhiss.news
14 regions plus AI & Technology, updated hourlyAboutKilwhiss GroupEditorial approachContactRSS

Researcher’s two million ‘canary’ prompts expose leaks across AI supply chain

·

A security researcher who sent almost two million AI prompts containing unique, traceable secrets has found widespread failures to keep user data private, Ars Technica reports.
The researcher, chief research officer at security firm Dragos, tracked where the planted secrets later appeared. He found signs that attackers had breached at least three AI providers and taken the prompts sent to them, and that several inference resellers had recorded prompts and handed them to others.
Several providers also failed to honour privacy flags such as zero data retention or no data collection.
The report explains that prompts, while protected by encryption on the wire, are opened up in plain text at every stage as they move through AI routers and pools of hosted models, and each party may log and store them under its own policies.

Source: Ars Technica

Kilwhiss News summaries are written in our own words and always credit and link to the original reporting. Read our editorial approach.

Latest headlines