How Bad Actors AbUse LLMs The Mechanisms That Enable Misuse and What We Can Do About
summary
Teams detail concrete defenses and failures: they use predefined prompts and OpenAI guardrails to constrain chat features, deploy Phoenix tracing and Grafana dashboards to detect anomaly clusters, and refuse to let LLMs give legal advice after seeing risky contract summaries. They prototype agents but pause agentized actions because hallucinations and content‑born prompt injection can trigger unwanted workflow changes. Recommended mitigations include least‑privilege access, reviewer attestations, kill switches, and strict CI/CD gates for AI‑generated code.
AI-written summary · about
chapters (10)
- 0:00Introduction to panel
- 3:18Observed AI misuse trends
- 7:10AI implementation decisions
- 10:44Regulated industry AI use
- 12:30Prompt injection defenses
- 16:45Underestimated LLM risks
- 18:52AI and MCP servers threat
- 22:05Handling AI-driven spam
- 24:12Vibe coding approach
- 26:04Closing remarks
related talks
26mWhen AI Gets Manipulated: Security Risks in LLM Applications
39mLLMs everywhere: The future is now and the past keeps repeating itself
13mNew Challenges in Software Quality Brought by GenAI and LLMs
44mI Abuse Your LLM Live with 10 Security Vulnerabilities
20mSecurity Threats in Modern LLM Applications
17mArchitecting LLM-Based Chatbots using free open-source LLMs
36mCompromising LLMs: The Advent of AI Malware
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges