https://curasec.metacog.co.kr/tags/differential-privacy/Recent content in Differential-Privacy on CuraSecHugoen-usMon, 31 Aug 2026 19:07:02 +0000https://curasec.metacog.co.kr/insights/2026-08-31-memorization-is-not-extraction-tight-differential-privacy-bo/Mon, 31 Aug 2026 19:07:02 +0000https://curasec.metacog.co.kr/insights/2026-08-31-memorization-is-not-extraction-tight-differential-privacy-bo/<ul> <li><strong>Engineer — Learn:</strong> If you rely on DP guarantees to protect training data in ML pipelines, this research shows that controlling memorization and controlling extraction are formally separate — a model can be memorized yet unextractable, or vice versa. Revisit your threat model assumptions, but no system change is required today.</li> <li><strong>SOC/IR — Skip</strong></li> <li><strong>Leader — Skip</strong></li> </ul>https://curasec.metacog.co.kr/insights/2026-08-31-revisiting-continuous-noise-sampling-for-multi-party-differe/Mon, 31 Aug 2026 19:07:02 +0000https://curasec.metacog.co.kr/insights/2026-08-31-revisiting-continuous-noise-sampling-for-multi-party-differe/<ul> <li><strong>Engineer — Learn:</strong> The paper exposes a fundamental flaw in sample-and-scale DP noise protocols, achieving near-100% membership-inference success against Orchard and DP-BREM+; engineers building federated analytics or DP aggregation pipelines should audit whether their noise-sampling implementation uses the vulnerable scaling approach.</li> <li><strong>SOC/IR — Skip</strong></li> <li><strong>Leader — Skip</strong></li> </ul>