AI Agent Liability Is a Huge Market – Insurers Are Fleeing It

01 /The Concept

Klaimee is building insurance for AI agent professional liability.

The startup is currently in Y Combinator, in the risk assessment phase. Actual policy issuance with monetary coverage is the next step – and the startup says it's coming soon.

Klaimee's initial focus is startups selling AI agents to enterprise clients – where having a policy in place can actively accelerate the sale. Potential damages from a misbehaving AI agent would be covered, which reduces the buyer's perceived risk and removes a common objection.

A parallel market is established companies using AI agents – either self-developed or licensed from third parties – to deliver services to their own clients. These companies can get a single policy covering all their AI agents, regardless of vendor, without requiring each supplier to carry separate coverage.

Initial risk assessment takes 3–6 weeks, followed by mandatory ongoing monitoring – whose cost is naturally built into the premium. This transforms insurance from a periodic one-time purchase into a continuous subscription-based relationship. For the insurer, that's obviously the better business model.

Klaimee's risk assessment methodology covers five dimensions.

Human oversight: how humans monitor the AI agent's actions, which errors are escalated to human review, what must be explicitly approved, and what can be overridden.

AI architecture: which model is used, how it's adapted to the company's context, what constraints and guardrails are built in, how well-designed the prompts are, and how the AI's outputs are automatically checked and filtered.

Technology stack: security of the underlying infrastructure, authentication and data transfer design, monitoring and logging setup, and deployment process for new agent versions.

Boundaries of responsibility: what systems the agent can access, what actions it can and cannot take, what categories of decisions it can make autonomously, and how human approval gates work for critical operations.

Failure behavior: known hallucination patterns, behavior in edge cases and rare scenarios, whether users can modify agent prompts, and what happens when compute resources are constrained or the underlying model provider goes down.

The assessment produces a score from 0 to 100. Above 70 earns an insurance certificate, with premium pricing inverse to the score – the higher the score, the lower the cost. Scores between 55 and 69 can still qualify, but with additional conditions, mandatory human involvement in specified situations, and explicit carve-outs from coverage.

As a side note on Crosby: its AI outputs are always reviewed by licensed attorneys before delivery to clients. It's plausible this approach – where a human formally signs off on the result – may allow the startup to qualify under standard professional liability insurance rather than needing AI-specific coverage. Though there's still room for a claim to find a foothold.

02 /Why It Matters

There's a category of insurance called Errors & Omissions (E&O), also known as professional liability insurance – typically carried by companies that provide professional services.

These policies cover potential damages when clients claim that a firm made mistakes, delivered incomplete work, or failed to deliver services at all. In many jurisdictions, E&O coverage is a mandatory licensing requirement for certain professional services – medicine, law, accounting, architecture, construction, and even real estate brokerage.

The professional liability market is already large and growing. Global market size was estimated at between $6.5 billion and $16.8 billion in 2025 depending on the source and scope, with projections of $31 billion by 2031.

A popular trend right now is startups repositioning themselves as professional services firms – delivering outcomes using their own AI platforms, rather than selling platform subscriptions. Delivering a service with your AI is a much more defensible and profitable business than selling access to the same AI.

To be taken seriously in regulated professional markets, some of these startups are registering as licensed professional services providers and obtaining the necessary E&O coverage. Crosby ([related review](/review/za-283-dnja-do-400-millionov)), which provides AI-powered legal services, reached a $400 million valuation within 283 days of launch by doing exactly this.

Seemed like the right call. But as it turns out, AI-powered companies are hitting an unexpected wall.

Traditional insurers are quietly excluding AI-caused errors and omissions from their standard professional liability policies.

They're getting this done legislatively where possible – in the US, state-level regulatory approvals are being secured. Even the ISO standard policy, which covers roughly 70% of the US insurance market, has made AI exclusion the default option – it kicks in unless explicitly overridden in the policy. The EU is adopting equivalent provisions starting this August.

This mirrors what happened with cyber insurance about 20 years ago. Traditional carriers began excluding cyber incidents by default – and an entirely new insurance market emerged to fill the gap. The global cyber insurance market now stands at $14–20 billion and is projected to reach $100 billion by 2032.

Given that AI will soon be involved in delivering nearly every type of professional service, the market for AI liability coverage could ultimately dwarf even the cyber insurance market.

03 /Opportunities

Every company deploying AI agents knows that AI agents make mistakes – which is why they build software safeguards to catch them.

A growing number of startups now sell monitoring and validation platforms specifically designed to continuously check AI agent performance for correctness.

These companies are one step away from something bigger: accepting financial liability for the reliability of that monitoring. Once they cross that line and structure it correctly, they become professional liability insurers. Why not take that step?

Or: enter this emerging market from scratch. It has every sign of becoming very large. Starting point: do what Klaimee has already started doing.