From Text to Flaws: Application Vulnerabilities with Generative AI and LLMs
summary
Explains how large language models (LLMs) and generative AI are integrated into applications, the typical vulnerabilities these integrations introduce (including concrete exploit stories), and practical countermeasures developers can implement to protect production systems. Covers LLM operation, common attack vectors against apps using generative AI, and developer-focused defenses and best practices.
AI-written summary · about
chapters (16)
- 0:00LLM introduction
- 3:54How LLMs work
- 7:49Building apps with LLMs
- 11:44Architecture vulnerabilities
- 13:09Prompt injection examples
- 16:55Data exfiltration via tools
- 18:17Code execution risks
- 19:20Image and text attacks
- 20:46Custom JPT leaks
- 21:51Classic injection protections
- 25:43LLM-specific attacks
- 31:24Dual LLM isolation
- 35:00User validation for ops
- 36:07Canaries and detection
- 38:14Best practices summary
- 39:15Conclusion and resources
related talks
46mFrom Text to Flaws: Vulnerabilities in Apps Using Generative AI and LLMs
40mFrom Text to Exploits: Vulnerabilities of Apps Using Generative AI and LLMs
42mFrom Text to Flaws: Application Vulnerabilities with Generative AI and LLMs
51mFrom Text to Flaws: vulnerabilities in applications with Generative AI and LLMs
31mHacking with Words...
45mOverview of Risks, Vulnerabilities and Paths to Safer Use of Generative AI
47mCurrent Security Threats in Large Language Model Applications
26mWhen AI Gets Manipulated: Security Risks in LLM Applications