The Mechanisms That Enable Misuse and What We Can Do About It
summary
Shows concrete exploits: a GitHub‑action CI/CD flow that uses Google Gemini for auto-triage and leaks tokens via prompt injection, proof‑of‑concept returning fake keys and a demo that works; measured effects like ~4.5‑hour active malware window, ~2‑hour median infection, 10% infected vs 99% exposed, and 20–60% hallucinated package suggestions enabling hallucination squatting and AI‑generated malware targeting dependency ecosystems and text-to-SQL style tool chains.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:53Prompt injection basics
- 8:32Tool and agent risks
- 13:50CI/CD pipeline risks
- 15:35Hacking Google
- 23:03Supply chain attacks
- 25:27Hallucinations and risks
- 31:10Malware speed limits
- 34:23AI malware threats
- 35:04Vong apocalypse and audits
- 36:51Insider threats and models
- 39:43Q&A
related talks
16mHarnessing AI for Malevolence: A Hackers Tale
34mHijacking Google’s CI/CD Through Prompt Injection:The New Era of AI-Based Exploits
51mPrompt Injection: When Hackers Befriend Your AI
37mSponsored Session: The Simple, Yet Lethal, Anatomy of a Software Supply Chain Attack
36mExploiting Vulnerabilities and Flaws to Attack Supply Chain
33mAgentic ProbLLMs: Exploiting AI Computer-Use and Coding Agents
32mBreaking the Chain: Advanced Offensive Strategies in Software Supply Chain
26mHow can we abuse AI hallucinations to feed the already bloated software supply chain