Demo · 57 seconds

Halt and release: your security tool stops one agent, a person lets it go

When an agent trips a rule, a Splunk alert can halt that one agent, the record shows what was enforced, and only a named person can release it.

Halt and release, 57 seconds. A local demo with fictional data. Splunk calls the halt API directly, not through a packaged connector. The record shows Partially enforced, with unresolved outcomes left visible. The demo does not reverse completed actions and is not a qualification pass.

Read the transcript

Your security tools just flagged an AI agent. It’s already on hold. With Decision Terrain, the tools you trust can stop an agent.

Your agents work in your systems, under your rules. When one tries something it shouldn’t, it’s blocked and logged. That event reaches your SIEM like any other security signal, and your alert fires. It requests a halt on that one agent, through a scoped API. New work stops.

The record shows who asked, when, which agent, and the real enforcement result, including what’s still unresolved. Only a named person can release it, with a reason. The halt and the release are both on the record.

Per agent. Human release. Recorded. That’s governed autonomy. That’s Decision Terrain.

On screen: “Does not reverse completed actions.” Fictional demo data, local development capture.

What happens in the demo?

The run follows one agent from a blocked write to a security hold and back to normal work, with a person making the final call.

An agent is working. Canary, an agent in a fictional defense-systems organization, is Ready on the agent list. It is given a task and moves to Working.
Policy blocks a write. Canary tries to write a restricted codename. The write is blocked and recorded, and the two denied events reach Splunk with the agent’s ID attached.
The alert halts that one agent. A Splunk scheduled alert calls the halt API through a scoped service identity. Canary moves from Ready to Security hold. The identity can target this agent only.
The record shows what was enforced. The intervention record names the requester, the accepted time, the agent, and the incident ID that matches the Splunk event. The result reads Partially enforced, with 13 unresolved outcomes left visible, and the record states that a halt does not reverse completed actions.
A named person releases it. Release needs a reason, the exact agent name, and acknowledgement of the unresolved outcomes. Canary returns to Ready, and the security ledger shows the halt and the release as separate events.

Why halt one agent instead of switching everything off?

An alert about one agent should not stop every agent. The halt is scoped to the agent named in the request, and the service identity that makes the call is limited to that agent. New work for that agent stops.

Release is a person’s decision. It needs a reason, the exact agent name, and acknowledgement of any outcomes that could not be confirmed, so a hold is not cleared by accident. Both the halt and the release stay on the record.

What the demo shows, and what it does not

The organization, agent, and data are fictional. This is a local development capture, not a customer deployment.
Splunk calls the halt API directly with a stored credential, through a local custom alert action. It is not a packaged Splunk connector.
The result is Partially enforced, with unresolved outcomes shown. The demo does not claim that previously admitted work was stopped, and a halt does not reverse completed actions.
The cut is edited and waits are removed, so it says nothing about how quickly an alert becomes a hold.
The halt record and the release are from one run, and the live status change is from another halt of the same agent. Another agent in the list was held earlier by a separate halt.
The agent ran on a hosted model. Nothing here is about self-hosted inference.
Security holds are deployment-gated. Confirm availability for your deployment.

Next steps

See a halt with your own security tools.

Walk through the hold and release with your agents and your SIEM, on your hardware or in your own cloud account.