Excluded Here. Affirmed There. Covered Nowhere?
A CFO opens two renewals on the same afternoon. The general liability form now carries a generative AI exclusion. The cyber renewal offers a new affirmative AI endorsement. She sets them side by side and asks the only question that matters: when the fall-prediction tool misses, which one answers the phone?
That's the market now. A nationwide review of state filings found 2,369 generative AI exclusion records in force on commercial general liability policies across 49 states by mid-2026 (Insurance Business).
The other side is moving just as fast. On September 17, Beazley confirmed affirmative AI cover in its cyber and tech E&O policies (Beazley). A week later it added two endorsements: one for organizations that shut down their own malfunctioning AI, and one for regulatory defense and penalties from unintentional misuse (Beazley). The names are AI Voluntary Shutdown and AI Regulatory Defence & Penalties. Beazley hasn't said where or when those two are available, so ask before you count on them. The same day, CFC added affirmative AI coverage to its intellectual property policy (Insurance Journal).
Affirmative feels like relief. It isn't, automatically. Navacord's Natalie Chan told Insurance Business that affirmative "does not necessarily mean better," and urged buyers to read "the actual insuring agreements, the definitions, the coverage trigger" (Insurance Business).
Call it the Split Market. The same AI failure can be written out on one line and written in on another. Which one responds depends on how the claim gets characterized.
And the shutdown cover asks a governance question before it asks an insurance one. Who in your organization is authorized to turn a malfunctioning AI off?
This month, run the 30-Day AI Coverage and Contract Audit. Ask your broker three questions and get the answers in writing:
- Has your carrier filed an AI exclusion endorsement in your states?
- Does your policy respond to a claim from an AI failure?
- Have you reviewed vendor contracts for indemnification language?
Then add a fourth: who can shut an AI system down, and does our cyber policy reward or penalize that decision?
Read the full research behind this story, plus this week's senior housing and governance flagships, in the LITT Briefing on Substack.