govern() · live agent decisions
AI agents request actions — the gate decides in real time
Every tool call an agent attempts is evaluated by govern() and
allowed, escalated to a human (step‑up), or denied
— then sealed into the tamper-evident audit chain. Click Run agents to stream a live batch.
ALLOW — routine reads
STEP-UP — money / PII (human approval)
DENY — destructive actions
Allow
2
routine reads permitted
Step-up
2
escalated for human approval
Deny
1
blocked by policy
Decision stream
live · 5 / 24h| Time | Agent | Action | Tier | Verdict | Reason |
|---|---|---|---|---|---|
| 12:04:21 | recon-agent | read:identity-graph | tier1 | ALLOW | allow-read |
source_system=AGT govern() harness
tenant=kpmgplayground
source_mode=scenario
agt=4.1.0
2026-06-14