Prompt Injection Firewall

…

Prompt injection against a SOC triage agent, run twice side by side. The left column has no defenses; the right has the ones you pick.

What the modes mean

replay — a real model run recorded earlier. The defenses re-execute on playback, so the toggles still work.
live — calls the model now. The same attack lands on some runs and not others.

Defenses for the right column: D1 Defense based on tags D2 Defense based on human review D3 Defense based on corporate context

What each defense does

D1 Defense based on tags: Mark the injected prompt as “suspicious” and leave LLM to decide whether to accept or reject the injected information.
D2 Defense based on human review: Have human decide whether to approve or disallow the injected information to be considered for the current run and memorized for future runs.
D3 Defense based on corporate context: By integrating with existing context, autonomous SOC can approve or reject the injected information without human intervention.

Undefended

Pick a scenario and press Run.

Same attack, defenses on.

Auto-update · automatically saved in long-term memory — a later run will recall it
Action-needed · held for review — if reject, a later run will not see it

Nothing saved yet.