AI Showdown: The Good, The Glitchy, and The Risky
summary
Compares specific frameworks and deployments, explaining training sources, Llama.cpp had a CVE 9.8 during model loading, and showing supply‑chain impact with ~250,000 attacks in 2023; evaluates cloud APIs versus on‑prem stacks like AMAR/AMA and CoverAgent, and developer libraries (LangChain4J, Quarkus LLM, Spring AI, JLama) against OpenSSF checks and OWASP LLM risks. Demonstrates exploit classes — Black Mamba worm generated by ChatGPT and prompt‑injection examples — and recommends containment, tests, and OpenSSF scoring for libraries before shipping.
AI-written summary · about
chapters (5)
- 0:00Introduction and AI overview
- 13:21AI benefits and real-world examples
- 23:08AI risks and security challenges
- 33:07Cloud-native AI and future trends
- 42:02Conclusion and recommendations
related talks
57mAI Applications Security Puzzle
54mA Practical Guide to AI Threats & Mitigation Strategies
35mGenAI Application Security: Not Just Prompt Injection
30mNavigating the New AI Security Landscape
06mData Privacy Concerns and Solutions When using LLMs in Production
53mBreaching LLM-Powered Applications: Overcoming Security and Privacy Challenges
01h 01mIntroduction to AI Security
37mAI Goat: A Damn Vulnerable AI Infrastructure