Limited Pilot · 4–8 weeks · fixed scope

Pilot Scope of Work

A fixed-scope engagement that puts pre-execution governance in front of your AI agent: it blocks reachable forbidden states before a tool runs. Clear deliverables each week, measurable success criteria, and a go/no-go gate — no open-ended consulting.

Start with a free assessment →Scope a pilot →
  1. Week 0

    Kickoff & scope lock

    • Mutual action plan: dates, owners, success criteria, exit gate.
    • Tool-manifest intake (or reuse your Day-1 assessment) → Ω exposure map.
    • Security & data-handling review: deployment topology, metadata-only logging, no raw payloads stored.
  2. Weeks 1–2

    Ω registry & corpus

    • Build your sector Ω registry — forbidden states mapped to your threat model (geometry unchanged, Ω expands).
    • Assemble a labelled evaluation corpus (adversarial + benign) from your real trajectories.
    • Baseline: which catastrophic states are reachable today, and which tools reach them.
  3. Weeks 3–5

    Integration in staging

    • Governance middleware in front of the agent's tool calls (evaluate-plan / evaluate-step) in your staging environment.
    • ALLOW / BLOCK / ESCALATE wired into the agent's execution path — nothing runs until a verdict is returned.
    • Tune to 0 false-positives / 0 false-negatives on the corpus; human-escalation path for edge cases.
  4. Weeks 6–7

    Validation & evidence

    • Replay every verdict with an engine-commit + ruleset-hash attestation (reproducible, tamper-evident).
    • Latency benchmark on your hardware (engine compute is sub-millisecond).
    • Board- and auditor-ready evidence pack: coverage, corpus results, replay proof, audit trail.
  5. Week 8

    Go / no-go gate

    • Readout against the agreed success criteria.
    • Clear decision: proceed to production + advisory retainer, or stop with a documented exposure report.
    • Production rollout plan and ownership if proceeding.

Success criteria

  • 0 false-positives / 0 false-negatives on the agreed evaluation corpus.
  • Every blocked trajectory is replayable and attested (engine commit + ruleset hash).
  • Sub-millisecond engine evaluation, measured on your hardware.
  • A named owner and a documented path to production governance.

Who does what

  • You provide. Tool manifest + sample real trajectories · a staging environment · an executive sponsor · a few hours/week of an engineer who knows your tool surface.
  • We provide. The engine + governance middleware · your bespoke Ω registry · corpus construction · integration support · the attested evidence pack.

Indicative investment: £250K–£750K+ depending on agent complexity, integration surface, and sector. Founding design partners receive preferential terms. The 48-Hour Runtime Governance Audit (£40K–£75K) is a faster, lower-commitment entry point that de-risks the pilot.

Indicative engagement scales only. Final commercial terms are determined following assessment and deployment review. The “+” symbol denotes that figures are not ceilings.

Assess your agent — free →All engagement options →