What it is
Defense-in-depth for the AI agents that now run critical infrastructure and essential facilities. THIRE applies seven independent control layers to every agent action and attests the result on a tamper-proof ledger.
Buildings are becoming agentic. The newest attack surface in critical infrastructure is not a network port — it is obedience without a check. THIRE makes agent behavior bounded, observable, and provable, so an accountable executive can ship an agent into the physical plant without shipping an unguarded hand.
Who it is for
The buyer is the executive accountable for critical infrastructure and essential facilities — power, water, transport, healthcare, the floor that cannot go dark. Commercial real estate and industrial facilities are the built-environment subset; the protected object is the agent, and the consequence is physical.
Seven control layers
- Identity & capability scoping
- Tool & command execution control
- Network egress containment
- Secret & credential isolation
- Memory & context integrity
- Human-in-the-loop authorization
- Escape prevention & runtime attestation
Each layer is independent; a failure in one must not defeat the others (fail-closed). See modules.
Three duties
ActiveDefense
guard
Stand at the line. Deny-by-default, fail-closed. Red means stop.
ActiveAI
govern
Keep the agent inside its job — scope, context integrity, human signature.
ActiveVuln
reason
Reason about exposure and cost, on the record. Evidence, not assertion.
Status
This is a read-only product face. Endpoints remain placeholder where no published contract exists (see the honesty note). Evidence labels apply to every claim; the default verdict is HOLD. No scanning or execution happens from this site — it forwards intake only.
Discovery
- llms.txt · llms-full.txt — agent-readable product text
- discovery.jsonld — Schema.org dataset
- intake contract — governed, loopback-only
- agent card · security.txt