For carriers, MGAs and reinsurers
The exclusions are already being filed
AIG, Great American and WR Berkley have all gone to regulators for AI exclusions. The Berkley language reaches any actual or alleged use of AI, including products that merely incorporate it.
On the record
-
AIG, Great American and WR Berkley have filed with US regulators to exclude AI-related risk from corporate policies
techcrunch ↗ -
The WR Berkley language reaches any actual or alleged use of AI, including products and services that merely incorporate it, drafted for D&O, E&O and fiduciary
insurance business ↗ -
AIG told regulators it has no immediate plans to use its exclusion, but wants it available as claim frequency and scale rise
techcrunch ↗
The stated reason is not what you would expect
One company’s AI deployment going wrong is a loss an insurer can absorb. A thousand or ten thousand of them, all at once, from the same underlying failure, is not.
The blocker carriers describe is correlation, not severity. A single bad deployment is underwritable. What cannot be underwritten is not knowing whether the failure inside one insured is sitting inside every other insured running the same pattern.
That is a question about failure patterns, which is the thing we collect
We attack live AI systems on purpose and record what broke, across many deployments, independent of the insured. It produces the evidence an underwriter would need to tell a one-off from a systemic exposure.
- Whether it is correlated
- the same failure pattern, tested across many deployments, so you can see whether a hit here means a hit everywhere
- What actually failed
- which control gave way under attack, rather than which controls the insured reported having
- What got reached
- the boundary that was crossed and what sat behind it, including whether it was a boundary between customers
- Whether it moved
- the same system re-tested after each deployment, so posture is a series rather than a submission-day snapshot
Worth a conversation before the exclusion becomes the default
We are not underwriters and we do not price risk. We generate and structure the failure evidence that pricing it would require.