How to Red Team AI Like a Pro: OWASP GenAI Manual Explained
カートのアイテムが多すぎます
カートに追加できませんでした。
ウィッシュリストに追加できませんでした。
ほしい物リストの削除に失敗しました。
ポッドキャストのフォローに失敗しました
ポッドキャストのフォロー解除に失敗しました
How to Red Team AI Like a Pro: OWASP GenAI Manual Explained
-
ナレーター：
-
著者：
Don't just ask if an AI model can be tricked into saying something bad. Ask what an attacker could actually achieve through it, whether you would notice, and whether your fix actually reduces the risk.In this episode, I walk through OWASP's GenAI Red Teaming Manual Version 1.0 (October 2026) step by step — a practical, repeatable process for adversarial testing of generative AI models, applications, and agents. Especially relevant if you work in banking, risk, or financial services, where AI touches money, customer data, and regulated decisions.What you'll learn:- The 8-phase OWASP red teaming process, from planning to remediation- How to scope a test properly, so findings lead to decisions, not just a long list- How to map an AI system's attack surface, including tools, retrieval, and other agents- Direct vs. indirect prompt injection, and why a banking assistant example turns a wording trick into a data breach- Persistence: how a compromise can survive into a later session- How to score impact and blast radius, and what to contain first- How to write a report that leads to fixes, with a clear owner per finding- How to fix in layers, retest, and turn attacks into detectionsSource:OWASP GenAI Security Project, GenAI Red Teaming Manual Version 1.0, October 2026