Skip to content
Models · Agents · Autonomy

The first liabilityclass where thesystem decides

An AI system does not fail like software. It fails probabilistically, in the tail, and it acts. When a model is given tools, budgets and the authority to execute, the exposure moves from professional indemnity into something closer to operational liability. 1B underwrites that shift.
Risk class
AI Risk
Focus
Models · Agents · Autonomy
Related products
NOVA
Operating model
Verify → Monitor → Prevent → Recover → Insure
Model behaviourIDLE
ACTIVATION · DRIFT · TAIL EXPOSURE · AUTONOMY LEVEL
The gap

Why the traditional market struggles

Technology E&O assumes a defect that can be traced, reproduced and patched. Model behaviour is emergent, distributionally sensitive and often non-reproducible. Causation is contested, the chain of responsibility spans model developer, deployer and integrator, and the loss can be reputational, regulatory and financial at once. Underwriting it requires evaluation evidence, not assurances.

Exposure surface

What we structurearound.

The risk surfaces that shape the programme. Cover is structured per counterparty, this is the map, not the wording.

Erroneous output and decision harm

Financial or physical loss caused by model output relied upon in a live process.

Autonomous agent action

Exposure created when a system holds tool access, spend authority or the ability to transact.

Model integrity

Data poisoning, prompt injection, model extraction and adversarial manipulation.

Intellectual property and content

Third-party claims arising from training data, generated output and downstream distribution.

Regulatory and compliance

Exposure under emerging AI regimes, including obligations attaching to high-risk deployments.

Performance and availability

Degradation, drift and dependency failure in production systems on which clients rely.

How 1B operates here

The loop, appliedto this class.

The same five stages, calibrated to the specific evidence this risk class produces.
  1. Verify

    Review of evaluation methodology, guardrails, human-in-the-loop design and deployment boundaries.

  2. Monitor

    Continuous signal on drift, incident rate, autonomy scope and change in model or provider.

  3. Prevent

    Agreed containment: capability limits, escalation triggers and rollback conditions.

  4. Recover

    Incident response covering technical containment, disclosure and third-party exposure.

  5. Insure

    Risk transfer calibrated to the autonomy level actually operating in production.

Continuous signals

What stays under observation once the risk is bound.

  • Autonomy level in production
  • Evaluation and red-team cadence
  • Human oversight coverage
  • Tool and spend authority scope
  • Model and provider change rate
  • Incident and near-miss telemetry
  • Data provenance posture
Built for
  • AI model developers
  • Enterprises deploying agentic systems
  • Healthcare, legal and financial AI vendors
  • Platform and infrastructure providers
  • Investors underwriting AI portfolios
AI Risk

This risk isalready live.The insuranceshould be too.