Humans sign. Agents assemble.
A five-tier workforce that keeps a named executive accountable for every regulated output — while agents do the continuous, evidence-heavy work underneath.
You choose the AI. Cabier orchestrates the trust.
The workforce chart
Five tiers. One accountable human per output.
The chart is intentionally shallow. Every regulated output has a Tier 0 signer at the top and a chain of agents underneath — each with a scope, a model dependency, and a kill switch.
Tier
Tier 0 · Accountable Human
Role
Named Senior Manager / Board sponsor
Scope
Signs every material output. Owns the regulatory answer. Cannot be delegated to an agent.
Examples
- · Chief Risk Officer
- · Chief AI Officer
- · Head of Compliance
- · SMF holder (UK)
Tier
Tier 1 · Orchestrators
Role
Human-led orchestration bench
Scope
Cabier practitioners and client leads who scope work, review agent output, and route the evidence trail to a Tier 0 signer.
Examples
- · Engagement Partner
- · Delivery Lead
- · Regulatory Reviewer
Tier
Tier 2 · Assurance Agents
Role
Read-only, evidence-anchored
Scope
Continuous control monitoring, twin-vs-twin diffs, KEV watch, ORS telemetry. Never mutate client systems. Every finding cites a signed audit ID.
Examples
- · CCM agent
- · Twin diff agent
- · Evidence collector
- · Enforcement watcher
Tier
Tier 3 · Draft Agents
Role
Human-in-the-loop, sandboxed
Scope
Draft board packs, regulator briefs, control narratives, remediation tasks. Output is queued for a Tier 1 reviewer and a Tier 0 signer — never auto-released.
Examples
- · Board pack drafter
- · Regulator brief drafter
- · Policy diff drafter
Tier
Tier 4 · Gate Agents
Role
Trust Gate enforcement
Scope
Machine-executable gates that block a workflow when a live obligation drifts. All actions are logged, reversible, and referable to a named executive.
Examples
- · Trust Gate blocker
- · Attestation gate
- · Reserve-freshness gate
Control envelope
What every agent runs inside.
Every agent is observed
Every agent invocation produces a signed audit ID (agent, model, prompt hash, evidence set, human reviewer, signer). No shadow calls, no unattributed output.
Model-agnostic by design
Agents are wired through the CABIER AI control plane. The client chooses the model; Cabier orchestrates the trust. Swapping GPT-5 for Claude 4.5 or a sovereign small model changes nothing in the accountability chain.
Nothing autonomous ships
Tier 2 is read-only. Tier 3 drafts land in a review queue. Tier 4 only enforces gates a human has pre-approved. No agent can sign, publish, or transact without a named executive.
Kill switch is real
Every agent tier has a per-domain and global kill switch owned by a Tier 0 executive. Trust Gates continue to block execution even when the drafting agents are paused.
Guardrails