Insights / AI liability

AI Liability: New Technology, Familiar Claims

AI is mostly a new way of causing familiar liabilities. The change comes with agentic systems, which can plan, decide and act without human sign-off, making losses faster, larger and harder to attribute.

Where the exposures sit

• Incorrect outputs and hallucinations relied on by customers or third parties.
• Intellectual property claims over training data and outputs.
• Privacy and data protection breaches.
• AI agents taking actions without human approval.
• Faster, more convincing cyber attacks.
• D&O exposure from weak oversight of AI or overstated AI capability ("AI washing").

These claims will mostly arise under existing policies (general liability, professional indemnity, cyber, technology E&O and D&O) rather than standalone AI cover.

The claims analysis stays the same

Duty, breach, causation, foreseeability and quantum all still apply. The new challenges are allocating responsibility between the developer, the vendor and the business deploying the system, and securing the evidence: logs, prompts, outputs, model versions and records of human review. A business using AI may remain liable to its customers even where the system came from a third party, so contractual indemnities and recovery rights against vendors matter.

Wordings are inconsistent

There is no consistent market approach. Many policies are silent on AI, some give affirmative or sublimited cover, and some exclude it. That uncertainty over which policy responds is likely to drive coverage disputes.

Implications for the market

Aggregation is the main concern. A single defective model or update could generate many related claims across a large number of insureds at the same time, testing event definitions and aggregation clauses across both primary and reinsurance programmes.