APCby ALEPH LABS
ACTION SECURITY FOR AI SYSTEMS / EARLY EVALUATION

Let agents move fast.
Keep actions governed.

APC is an action-security project from Aleph Labs exploring how authorization decisions, policy state, action details, and denial behavior can stay connected across an AI-agent workflow.

Enter the playground Understand the scope
Opening the playground in a few seconds…
● Browser-only demo● Synthetic scenarios● No external actions
APC / ACTION FLOW SIMULATED
01Agent intentRequested actionINPUT
bind exact action
◇Security checksPolicy · freshness · targetCHECK
synthetic decision
✓ALLOWWithin demo rules
×DENYRule mismatch
✓Demo only: no provider is called and no action is executed.
01Action-awareInspect the requested action and target
02Fail-closed examplesSee why a synthetic scenario is denied
03Visible decision traceReview each simulated check
A WALKTHROUGH, NOT A CLAIM OF PRODUCTION READINESS

Make the decision
legible.

Start with a synthetic scenario, change a control, and examine how the example decision changes. The goal is to make action-security requirements easier to discuss, test, and prioritize with engineering and security teams.

◎
CHECK 01

Bind the action

Keep the principal, requested operation, target, and relevant parameters visible together in the example decision.

◇
CHECK 02

Test the guardrails

Explore synthetic denials for stale authority, policy drift, replay signals, unknown principals, and out-of-scope targets.

⌁
CHECK 03

Read the trace

See which illustrative checks pass, which check blocks the request, and what reason is attached to the result.

READY TO TEST A SCENARIO?

Open the synthetic playground.

Try the default safe-read scenario, then switch to a denied-action case.

Launch playground
TRANSPARENCY FIRST

What this preview is.
And what it is not.

✓ INCLUDED IN THIS PREVIEW
  • Browser-only, deterministic synthetic scenarios
  • Illustrative allow / deny outcomes and reason codes
  • Local decision traces using fabricated data
  • A starting point for bounded evaluation conversations
! NOT PROVIDED BY THIS PREVIEW
  • No connection to an APC enforcement backend
  • No live identity provider or external policy decision point
  • No cloud, SaaS, payment, or infrastructure action execution
  • No proof of certification, production readiness, or universal enforcement

APC's deployable evaluation candidate is separately limited to demo and lab use. Live sensitive-action enforcement requires additional, action-specific integration, verification, and assurance work. This public playground must not be used to authorize or perform real actions.