Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.
Summary
A security researcher embedded two million unique, trackable 'canaries' (secret data fragments) into the AI supply chain. Testing prompts with these canaries revealed that AI models are leaking this sensitive data, indicating a significant privacy concern within AI systems.
IFF Assessment
FOE
The research highlights how AI models can unintentionally leak sensitive data, posing a risk to privacy and security.
Defender Context
This research underscores the potential for AI models to expose sensitive information that is inadvertently included in their training data or prompt inputs. Defenders should be aware of the risks associated with data leakage from AI systems and implement appropriate controls and monitoring.