https://peertube.eqver.se/videos/watch/a61875ad-ca30-4221-ae08-d8b29819d811 A DEF CON 32 talk where security researcher Suha Hussein presents a new class of attacks on machine learning systems. Instead of targeting only the model, attackers exploit flaws in input processing within ML pipelines to stealthily introduce backdoors. The talk demonstrates how vulnerabilities in model serialization and specially crafted files can be used to compromise ML systems. It also provides recommendations for securing ML pipelines against such attacks. Fri, 24 Jul 2026 11:26:01 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://peertube.eqver.se https://peertube.eqver.se/client/assets/images/icons/icon-512x512.png https://peertube.eqver.se/videos/watch/a61875ad-ca30-4221-ae08-d8b29819d811 All rights reserved, unless otherwise specified in the terms specified at https://peertube.eqver.se/about and potential licenses granted by each content's rightholder.