Why AI Agent Insurance Will Be a $10B Market

Insurers are circling the AI agent opportunity. But they cannot price what they cannot measure. The market that solves the evidence problem wins the decade.

The signals are everywhere

Three moves in the past twelve months tell the same story: institutional capital believes AI agent liability is an insurable risk, and the first movers are positioning now.

The actuarial void

Every insurance product in history has been priced on loss data. Fire insurance drew on centuries of building fire records. Motor insurance was built on decades of accident statistics. Cyber insurance, the most recent category, still took years of breach data before underwriters could construct credible models.

AI agent insurance has none of this. There is no historical loss data for autonomous software agents making decisions, calling APIs, moving money, and modifying infrastructure on behalf of organisations. The actuarial tables are blank. Fire insurance has 300+ years of loss data, motor insurance has 100+ years, cyber insurance has 15+ years. AI agent insurance has zero.

This is not a temporary gap. It is a structural problem. Without a standardised way to record what agents do, how they were governed, and what happened when things went wrong, the data will never accumulate in a form underwriters can use.

The telematics parallel

Motor insurance faced a version of this problem in the early 2000s. Insurers had aggregate accident statistics, but they could not distinguish a careful driver from a reckless one until after the claim. Young drivers were penalised uniformly. Safe drivers subsidised dangerous ones.

Telematics changed everything. OBD-II dongles and smartphone sensors gave insurers real-time behavioural data: braking patterns, cornering speed, time-of-day driving, phone usage. Suddenly, underwriters could price individual risk based on observed behaviour rather than demographic proxies.

AI agent insurance needs the same transformation. Not demographic proxies (which LLM, which vendor, which industry), but direct behavioural evidence: what the agent actually did, what governance controls were active, and whether the operator can reconstruct any incident from first principles.

The analogy is precise. OBD-II gave auto insurers behavioural telemetry. VOLT gives AI insurers behavioural evidence. The operator who can prove their agent followed policy is the one who gets the better premium.

What underwriters actually need

Conversations with insurance professionals and analysis of early-stage AI insurance products reveal three non-negotiable requirements. Every underwriter asks the same questions, regardless of whether they frame them in actuarial language or plain English.

Sizing the market

Conservative modelling puts the AI agent insurance premium pool at £1.2 to £2.1 billion by the end of the decade, assuming 3 to 5% penetration of AI agent operational spend. When you add the surrounding ecosystem (evidence infrastructure, compliance tooling, certification services, claims management platforms, reinsurance), the total addressable market reaches £6.5 to £8 billion.

Adverse selection and the evidence gap

Insurance markets fail when they cannot distinguish good risks from bad ones. This is the adverse selection problem, and it is acute in the AI agent space.

Without standardised evidence infrastructure, insurers face a binary choice: insure everyone at high premiums (pricing out responsible operators) or insure no one (leaving the market to self-insurance and contractual liability caps). Neither outcome is efficient. Neither builds a market.

The operators who invest in evidence infrastructure, proper governance controls, and incident reconstruction capability are exactly the ones insurers want to underwrite. But without a way to verify those investments, the careful operator pays the same premium as the reckless one.

The evidence substrate

This is the problem VOLT and AOCL were designed to solve. Not as an afterthought or a compliance add-on, but as the foundational infrastructure layer that makes AI agent insurance viable.

VOLT provides the evidence chain: every agent action, every API call, every decision recorded in a tamper-evident hash chain. If an entry is modified or deleted, the chain breaks. The evidence is not an assertion by the operator. It is a cryptographic fact.

AOCL provides the governance proof: an 11-layer pipeline that enforces spending limits, human approval thresholds, scope restrictions, and safety checks. Every decision passes through the pipeline, and every governance verdict is recorded in VOLT. The underwriter does not need to trust the operator's claims about their governance posture. They can verify it.

Together, they satisfy all three underwriter requirements. Evidence of actions (VOLT chains). Proof of governance controls (AOCL verdicts in VOLT). Incident reconstruction (full chain replay from any point).

The SOC 2 trajectory

Consider the parallel with SOC 2 compliance. Twenty years ago, cloud providers competed on features and price. Security posture was invisible. Then SOC 2 emerged as a standardised framework: auditors could verify controls, and customers could compare providers on a common basis.

SOC 2 did not just help buyers. It helped sellers. The providers who invested in security could finally prove it, differentiating themselves from competitors who merely claimed to be secure. Compliance became a competitive advantage, not just a cost centre.

AI agent evidence infrastructure is on the same trajectory. The operators who adopt VOLT-compatible evidence chains now will be the ones who get insurable first, at the best rates. As the market matures, evidence infrastructure will shift from differentiator to table stakes, exactly as SOC 2 did for cloud infrastructure.

Certification tiers

To make evidence infrastructure legible to insurers, we are designing a tiered certification framework. Each tier represents a progressively stronger evidence posture, giving underwriters a clear signal for risk pricing.

Who wins this market

The AI agent insurance market will not be won by the insurer with the best marketing or the largest balance sheet. It will be won by whoever controls the evidence layer.

This is not speculation. It is the pattern every insurance category follows. The company that standardises risk measurement captures the market. Verisk did it for property/casualty with actuarial data. Fair Isaac did it for consumer credit with FICO scores. The telematics providers did it for usage-based auto insurance.

For AI agent insurance, the evidence substrate is the prize. The protocols that become the standard for recording agent behaviour, proving governance compliance, and enabling incident reconstruction will sit at the centre of a multi-billion pound market.

That is what we are building with VOLT, AOCL, and AEE. Not an insurance product. The infrastructure that makes AI agent insurance possible.