Reaction
Dragos says prompt traps exposed breaches at three AI providers
Dragos reports prompt theft through reseller logs and privacy failures, including ignored zero-data-retention settings.
TLDR
Using nearly two million prompts seeded with unique URLs and credentials, Dragos says it found evidence of three AI provider compromises between July and November 2026. Researchers report attackers used planted AWS keys, while some resellers collected prompts through built-in logs without an exploit. The firm also says some stolen prompts carried zero-data-retention flags, and that several providers have begun investigations.
Combined views
1.1K
4 Sources, first seen ago
Comments