Poison Once, Compromise Many: How Model Reuse Amplifies AI Vulnerabilities @socallinuxexpo
Poison Once, Compromise Many: How Model Reuse Amplifies AI Vulnerabilities @socallinuxexpo
Uploaded April 2026 | Updated September 2026, 1 week ago
Talk by Jason Kramer

socallinuxexpo.org/scale/23x/presentations/poison-once-compromise-many-how-model-reuse-amplifies-ai-vulnerabilities

AI models are rarely built from scratch. Through model reuse and transfer learning, organizations inherit risks they may never see. This session explores how backdoors, poisoning attacks, and evasion techniques survive across generations of models, exposing downstream systems to compromise.
Poison Once, Compromise Many: How Model Reuse Amplifies AI VulnerabilitiesBarns Dont Burn Down Because of One Loose Nail, But Yours Just MightFrom Bash to Burnout: Staying Sane in a 24/7 Tech WorldCloudNativePG: Robust, Self-Healing PostgreSQL on KubernetesMark Russinovich: Toward a Secure and Sustainable Open Source Supply ChainHow We Achieved a 40x Speedup with Custom Codegen for OTTLData on Kubernetes / stateless storageBallroom A Saturday Mar. 07 - SCaLE 23xSpeeedy IoT dev w/ Agentic TUIA Practical Guide to Training a Small Language Model: Tokenizers, Training, and Real-World PitfallsPorous by DesignDoug Comer: Software Distribution Now And Then - Why And How The Internet Changed Everything
Southern California Linux Expo |

Poison Once, Compromise Many: How Model Reuse Amplifies AI Vulnerabilities

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER