Skip to content

Learn · Security

Security in the AI era is engineering work.

Not a checklist. A working track on the security surface AI introduces — and the OWASP-class issues it quietly worsens.

Modules

Six modules.

01

OWASP, refreshed for 2026

Top 10, but specific to the codebases AI is most likely to subtly miswire — injection vectors that hide inside prompt boundaries.

02

Authorization drift in generated code

How AI-assisted edits gradually loosen auth boundaries. How to catch it in review, and how to defend against it in design.

03

Prompt injection as RCE

Treating prompt injection as a code execution primitive. Real attack surfaces, real mitigations.

04

Secure tool-use patterns

How to design the tools a tool-using agent calls. Whitelists, idempotency, blast radius, audit trails.

05

Secrets and SSRF

Where AI agents leak credentials, when fetched URLs become SSRF, and how to draw the network boundary.

06

Threat modeling with AI

Using AI to expand a threat model — without taking its word for what's exploitable.