miatz™
Log inSit Chuun
Free concept playground · no account

Give the Agent Access

Grant an agent permissions, then see the blast radius light up.

All 90 playgrounds Join the free cohort

Give the Agent Access

AvailableFree

Grant an AI agent permissions. Each grant adds to the blast radius — how far damage spreads if a prompt injection turns it against you.

  • blast 1/5 · reversible
  • blast 1/5 · reversible
  • blast 2/5 · reversible
  • blast 3/5 · irreversible
  • blast 5/5 · irreversible
  • blast 5/5 · irreversible
  • blast 5/5 · irreversible
minimalblast 0 · 0 irreversible
'Just give it access' is a spreading blast radiusleast privilege is the only safe default

These grants sum to a blast radius of 0 (minimal), with 0 irreversible. Nothing granted yet. Least-privilege by default — the agent can look but barely touch. Keep grants this tight and add scope only when a task provably needs it.

Agent permissions aren’t one checkbox — they compound. Read is safe; delete, deploy, and spend are irreversible, and a single injected instruction can trigger any grant you’ve made. Grant the smallest set that still does the job, gate the irreversible scopes behind human approval, and never hand one agent delete + deploy + spend together.

Was this playground useful?

Give the agent access

Grant an AI agent permissions and watch the blast radius light up. Each grant adds to how far damage spreads if a prompt injection turns the agent against you. "Just give it access" isn't one checkbox — it's a spreading, measurable risk. Runs entirely in your browser (0 uploads, works offline).

How to use it

  1. Toggle the permission scopes you'd grant the agent.
  2. Watch the blast-radius meter and risk tier climb — read scopes are cheap, delete/deploy/spend are not.
  3. Read the reveal: your total blast radius, how many grants are irreversible, and the single worst scope.

What this clears up (the fundamentals)

  • Permissions compound — each grant adds to the blast radius; three medium grants can outweigh one big one.
  • Irreversibility is the real risk — read and edit can be undone; send, delete, deploy, and spend cannot.
  • Injection turns your grants into weapons — a single injected instruction can trigger any permission you've handed over.
  • Least privilege is the only safe default — grant the smallest set that still does the job, and gate the irreversible scopes behind human approval.

Where it's used

An AI-operations and security drill — the fastest way to feel why agent autonomy is a risk decision, not a convenience toggle. It's a miatz build-lab concept playable; learn to build it yourself.

FAQ

What is "blast radius"?

How far the damage spreads if a permission is misused — here, a 1-to-5 score per scope plus whether the action can be undone. The total across your grants is the agent's overall reach.

Why is irreversibility so important?

Because a reversible mistake (an edit) can be rolled back, but an irreversible one (a sent email, a deleted record, a payment) cannot. Irreversible grants deserve human approval before an agent can use them.

How does prompt injection relate to this?

An attacker who slips instructions into the agent's input can make it use any permission you've granted. The fewer and less powerful the grants, the less an injection can do.

Is anything uploaded?

No. The scopes and blast-radius math run entirely in your browser — nothing is transmitted or stored. Turn off your Wi-Fi and it still works.

Limits

A teaching model with a fixed set of scopes and blast values — real permission systems are finer-grained. The least-privilege instinct transfers directly.

Related

Part of the Demystify Playgrounds. Explore the rest from the Playgrounds home.

Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.

More playgrounds

Ninety playgrounds. Zero setup.

Every concept here is playable free, no account — and inside the program you learn to rebuild the machinery yourself.

Browse all playgrounds Join free