AI Security
Same Credentials, Same Blast Radius: Securing 1,500+ Agent Skills at iFood
Roughly 1,500 skills approved, 350 blocked, and the pipeline that decides: static, semantic, custom rules and composition.
Research, offensive testing, and production defenses for machine learning systems, LLMs, agents, and AI-enabled products.
Roughly 1,500 skills approved, 350 blocked, and the pipeline that decides: static, semantic, custom rules and composition.
Four recorded talks from iFood's first AI Security Meetup.
A historical deep dive into LangChain RCE, SQL injection, and mitigation.