Discussion between Mehdi Medjaoui and Troy Leliard,API Security.
summary
Claim: APIs remain the primary route into AI and therefore the main attack surface; practical attacks demonstrated include prompt injection and credential leakage via exposed endpoints. Tools and approaches mentioned: Longchain deployments, simple developer inspector tools, dynamic API fuzzing and runtime protection, and vendors like Sky ID for model extraction prevention; Akamai + Nvidia partner pushes GPU inferencing to the edge. Surprising figures: a consultancy report claims ~70% adoption while physical outages (squirrels ~40%) still break systems.
AI-written summary · about
chapters (15)
- 0:00AI and API security
- 1:10Attack vectors in APIs
- 4:37AI in production
- 6:04AI to production challenges
- 7:31AI for security vs fun
- 9:17API security testing
- 11:03Prompt injection risks
- 13:32Current solutions and delays
- 15:21Physical data center threats
- 18:13System downtime causes
- 19:58Quantum-safe cybersecurity
- 20:19Teaching attacks on LLMs
- 21:22Small language models
- 22:49Model extraction flaws
- 23:10Wrap-up and next steps
related talks
23mTraceable AI
25mThe New API Vulnerability Frontier
31mWhy securing models without securing APIs creates a false sense of control
24mSecuring AI Integration with API's
26mAPI and Application Security - The Future
31mAPI security
01mOn The Value Of API Security Focus
51mThe Next Application Security Frontier: AI-Ready API Defense