Survey: most insurers would trust AI only on repeatable work

Just 6% of respondents said they would trust a general-purpose AI model with high-consequence underwriting or claims decisions, versus 83% who would hand it repeatable work.

A Business Wire release on ISG research states that 83% of the global insurance market surveyed would let AI handle repeatable work, but 75% would only do so with an insurance-specific or governed model. Six percent said they would trust a general-purpose AI model to execute high-consequence underwriting and claims decisions. Sample size and methodology are not disclosed in the release.

The source's own statement, as linked above. Nothing beyond it has been independently checked by us.

This is not a measurement of what any single carrier has deployed. It is a stated preference from a vendor-commissioned survey, not an audit of production systems.

Underwriting leaders evaluating a vendor's AI tool should ask whether the underlying model is insurance-specific or general-purpose before extending it past repeatable, low-consequence work.

• Ask any AI vendor pitching underwriting or claims automation whether the model is general-purpose or insurance-specific.
• Restrict general-purpose model use to repeatable tasks until governance catches up.

Insurance leaders want human sign-off on AI's biggest calls