BEGIN:VCALENDAR
VERSION:2.0
PRODID:Data::ICal 0.24
BEGIN:VEVENT
DESCRIPTION:   'Title: Breaching the Frontier: A Hands-On Lab on AI-Native\
 n   Vulnerabilities\n   Tags: Adversary Village | Creator Event/Activity\n
    When: Friday\, Aug 7\, 11:00 - 12:55 PDT\n   Where: LVCCW Level 1 Hall 
 2 602 (Adversary Village) Hands-on Activity\n   Area - [1]Map\n\n   Descri
 ption:\n\n   For thirty + years\, hacking has meant finding flaws in code 
 people\n   wrote: memory corruption\, injection\, and logic bugs. AI syste
 ms fail\n   differently. Their vulnerabilities live in the weights or the\
 n   supporting fixtures. Caches that speed up inference\, embeddings that\
 n   supposedly anonymize data\, weights that execute code\, training corpo
 ra\n   memorization\, misalignment\, reward hacking and the structural\n  
  inability of a language model to separate instructions from data.\n   The
 se are some of the AI-native bugs. They exist because of how these\n   sys
 tems are built\, not because a developer made a mistake\, and no\n   tradi
 tional AppSec tool will find them. As enterprises deploy LLM\n   assistant
 s\, agents\, and RAG pipelines into the adversary's path\, this\n   bug cl
 ass is becoming the new attack surface.\n\n   This lab walks you through l
 ive exploitation and defenses in a\n   contained environment. Built on pee
 r-reviewed work from NDSS\, USENIX\n   Security\, IEEE S&P/SaTML\, CCS\, a
 nd ICLR (2023-2026).\n\n   We will cover attacks across Model Serving side
  channels\, prefix cache\n   attacks\, speculative decoding\, the AI suppl
 y chain\, dataset poisoning\,\n   sleeper-agent backdoors that survive saf
 ety training\, and trojaned\n   LoRA adapters. We will also have examples 
 of the classic prompt and\n   chat template injection bugs baked in CTF st
 yle.\n\n   Every lab is built on released artifacts: papers\, code\, datas
 ets\, and\n   open models. Attendees leave with a working understanding of
  modern AI\n   systems\, how AI-native vulnerability research works and a 
 method they\n   can run on their next engagement.\n\n   Who should attend:
  red teamers\, pentesters\, and security researchers\n   fluent in traditi
 onal exploitation who want the AI-native equivalent.\n\n   Prerequisites: 
 a laptop with admin access that can run Docker. A\n   MacBook Pro with 128
  GB of memory is preferred if you like to run\n   models locally.\n\n   Sp
 eakerBio:  Anto Joseph\, Lead Product & Infrastructure Security at\n   Eig
 enLabs\n\n   Anto Joseph leads product and infrastructure security at Eige
 nLabs.\n   His work focuses on AI-native vulnerabilities — the class of se
 curity\n   flaws that arise from how modern AI systems are built rather th
 an from\n   mistakes in application code — and on translating cutting-edge
 \n   academic security research into practical\, hands-on offensive\n   tr
 adecraft for red teamers and security researchers.\n\n   Links:\n       ad
 versaryvillage.org/adversary-events/DEFCON-34/Anto-Joseph/ - [2]https://ad
 versaryvillage.org/adversary-events/DEFCON-34/Anto-Joseph/\n   '\n\n   1. 
 #LVCCW_Level1_Hall2\n   2. https://adversaryvillage.org/adversary-events/D
 EFCON-34/Anto-Joseph/\n\n\n
DTEND:20260807T195500Z
DTSTART:20260807T180000Z
LOCATION:Adversary Village - LVCCW Level 1 Hall 2 602 (Adversary Village) H
 ands-on Activity Area
SUMMARY:Breaching the Frontier: A Hands-On Lab on AI-Native Vulnerabilities
END:VEVENT
END:VCALENDAR