rotative
the decoder · November 22, 2026

Cheap AI providers caught stealing keys, snooping on users

security researchers at dragos found major security flaws in ai services using 2 million test requests, allowing attackers to steal credentials and keys undetected for months.

security researchers at dragos exposed major security flaws in ai services using roughly 2 million test requests. they identified at least three compromised providers where attackers stole user data undetected for months. the researchers found that intermediaries and model marketplaces that routed requests emerged as a major risk. they recommend that organizations control ai request routing, check privacy settings, and plant canary tokens. the incident highlights the risks of using cheap ai inference providers.

read at source ↗

← rotative