DCSG1 - Autonomous and Exploitable: Breaking AI Agents before they break everything else - Aaron Ang
Plays from YouTube in privacy-enhanced mode — nothing loads from YouTube until you press play.
The hottest thing in AI right now is agentic systems — LLMs that browse the web, write and execute code, manage your email, call your APIs. They're being deployed faster than anyone understands the attack surface. Join Aaron as he presents a systematic teardown of AI agent security: how agents inherit trust they shouldn't have, how a single adversarial webpage can hijack a browsing agent mid-task, and how multi-agent pipelines create exploit chains that no single guardrail can stop.
Description from DEF CON.
Explained
The technology and terms in this story. Learn more
- TechnologyAI agentAn AI model that acts on its own: it plans steps, uses tools and takes actions, not just answers.
- TechnologyLLM · Large language modelAn AI model trained on huge amounts of text to understand and generate language, such as ChatGPT or Claude.
- AttackPrompt injectionHiding instructions in content an AI reads, so it follows the attacker instead of its user.
- ConceptExploitCode or a technique that turns a software flaw into something an attacker can actually use.