Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced.
A researcher successfully tracked the leakage of sensitive data by embedding two million unique identifiers, known as canaries, into datasets used to train AI models. The results demonstrate that proprietary information frequently migrates from private training sets into public model outputs. This experiment highlights a systemic failure in data privacy protocols, suggesting that once information is ingested by large-scale AI systems, it becomes difficult to restrict or remove from the resulting technology.
Covered by 2 sources · 3 articles
- AArs Technica↗Dan Goodin1d ago
- HHacker News↗paulis1d ago
- HHacker News↗Fe2O310h ago