Live CoreGuard Demonstration

The agent authenticated successfully.
It still wasn't allowed to do that.

A deployment agent holds valid credentials for a production database. It proposes DROP TABLE production_customers. Every identity check passes. This page runs that request through the real EVE CoreGuard evaluator and shows you what happens next — then runs a permitted query from the same agent with the same credentials.

Authentication establishes who or what the agent is.
Authorization determines what the agent is permitted to do.
CoreGuard enforces that authorization before execution — and produces verifiable evidence of the decision.

Run it

Pick a scenario. The verdict is produced by the deployed CoreGuard evaluator against the autonomous_agent_v1 policy pack — deterministic rules, no language model anywhere in the decision path.

What actually happened

✗ Blocked

Agent (authenticated)
↓
DROP TABLE production_customers
↓
CoreGuard
↓
BLOCK
✗
Database
↓
Signed evidence

✓ Allowed

Agent (same credential)
↓
SELECT COUNT(*) FROM production_customers
↓
CoreGuard
↓
ALLOW
↓
Database
↓
Signed evidence

Same agent. Same credentials. Different authorization. That is the distinction identity infrastructure cannot make on its own — a credential says the agent is who it claims to be, not that this particular action against this particular resource is permitted right now.

The enforcement path

AI agent
   ↓
Authenticated                     ← identity established
   ↓
Proposed action                   ← DROP TABLE production_customers
   ↓
CoreGuard                         ← deterministic policy evaluation
   ↓
ALLOW / BLOCK / MODIFY            ← no LLM in this decision
   ↓
Execution boundary                ← a BLOCK never crosses it
   ↓
Signed evidence
What this demo is. The database is an in-memory Python dictionary. There is no SQL parser and no SQL executor anywhere in this path — the SQL you see is a display label attached to a fixed, server-side scenario, and no text you send is ever interpreted. The governance decision, the policy pack, the rule identifiers, the signing and the verification are the real production components. CoreGuard governs a proposed action submitted to it; it is not a SQL proxy and does not sit on the database wire protocol.
Read: Authenticated Doesn’t Mean Authorized → EVE CoreGuard → Authority Lab → Verify any EVE record →