LLMs everywhere: The future is now and the past keeps repeating itself
summary
Demonstrates live demos against Claude and Gemini that turn rendered HTML, hidden text and downscaled images into working prompt injection exploits, shows an LLM-generated JavaScript port scanner and cites real malware and a ransomware case that used LLMs for lateral movement and payload creation. Argues agent risk when models get tools — agentic AI writing to inboxes or databases — and recommends sandboxing, strict input layout checks and a deliberate human-in-the-loop with actionable checks rather than blind approvals.
AI-written summary · about
chapters (19)
- 0:00Introduction and keynote setup
- 1:52About the speaker and topic
- 3:17Why AI is relevant today
- 5:07The AI hype and market dynamics
- 7:00Practical uses of LLMs
- 9:26AI as a new attack surface
- 11:12OpenAI CEO's warning on AI risks
- 12:17Prompt injection explained
- 14:04Live demo of prompt injection
- 17:41AI agents and security concerns
- 20:37Human oversight and automation risks
- 22:07AI in malware and ransomware
- 23:12Hallucinations and trust in AI
- 25:02Q&A session begins
- 28:34Jailbreaking and session persistence
- 31:04Communicating AI risks to management
- 33:13Why we repeat security mistakes
- 36:21Local LLMs and emerging threats
- 38:06Closing remarks and Q&A wrap-up
related talks
36mCompromising LLMs: The Advent of AI Malware
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
43mLLMs in a Hacker's Hands: From Prompt Injection to Data Leakage
09mLLM hacking is underrated
44mI Abuse Your LLM Live with 10 Security Vulnerabilities
53mDoors of (AI)pportunity: The Front and Backdoors of LLMs
20mSecurity Threats in Modern LLM Applications
26mHow Bad Actors AbUse LLMs The Mechanisms That Enable Misuse and What We Can Do About