Securely Launching to the Public
Ship agentic applications to real users without exposing your system to prompt injection, data leaks, or runaway costs.
Threat models, access control, and operational safety for agentic products
Building something that works locally is one thing. Putting it in front of real users — with real data, real adversaries, and real consequences — is another.
This chapter covers the security and operational concerns that matter most when you launch an agentic application to the public. Threat modeling, authentication, monitoring, and the ability to shut things down when they go wrong.
Threat Modeling for Agentic Applications
Identify the unique attack surfaces of agentic applications — from prompt injection to data exfiltration — before they become incidents.
Authentication, Authorization, and Access Control
Design access control that works when agents act on behalf of users — where identity, permissions, and delegation all get complicated.
Monitoring, Incident Response, and Rollback
Build the operational infrastructure to detect problems, respond to incidents, and roll back agent actions when things go wrong.