Skip to content

Agentic security / 2026

Follow the signal.
Break the boundary.

Agents act, delegate, and remember. Explore the trust boundaries behind their decisions through tools, state changes, and verifiable evidence.

30 missions · 10 threat categories

Scripted practice is unranked. Live runs require server model configuration and verified Google sign-in. Defender missions evaluate your controls directly. Code missions use a bounded virtual interpreter.

Ranked standings
0/30 ranked · 0/30 practice

Practice progress is stored on this device. Ranked completions come from your verified account.

Threat taxonomy: OWASP Top 10 for Agentic Applications 2026. Missions are independently authored training scenarios.