Your liability policy may have carved out AI

If your AI feature causes a loss and a customer sues, whether your liability policy responds now depends on a form number.

In January 2026 the Insurance Services Office, the standards body Verisk owns and whose forms most US commercial general-liability policies are built on, published a set of endorsements that exclude generative AI. Insurers can attach them at renewal. The fact that a Munich Re company launched a separate product this spring to cover “losses that some General Liability policies exclude” is a good sign some already have. The coverage you think you carry may read differently than it did a year ago, and the only way to know is to read the endorsement schedule.

What the new forms exclude

Three endorsements carry a January 2026 edition date, per the Big “I” Virtual University’s breakdown of the multistate filing:

Form	What it excludes
CG 40 47	Coverage A and B: bodily injury, property damage, and personal and advertising injury arising out of generative AI
CG 40 48	Coverage B only: personal and advertising injury arising out of generative AI
CG 35 08	Products/completed operations: bodily injury and property damage arising out of generative AI

The scope is wider than the form names suggest. CG 40 48 defines generative AI as “a machine-based learning system or model that is trained on data with the ability to create content or responses, including but not limited to text, images, audio, video or code.” Then it removes Coverage B for any “personal and advertising injury” arising out of it. Personal and advertising injury is the bucket that catches libel, copyright, and privacy claims from marketing output. Which is to say the exclusion lands squarely on the AI-written blog post, ad, or social caption a small team is most likely to ship without a lawyer reading it first.

The exclusion opened its own market

Underwriters do not usually cut coverage and leave the demand unanswered. Two products launched within a month of the exclusions taking effect.

Mosaic launched cover built on Munich Re’s aiSure platform on February 26, 2026, offering up to EUR/USD/CAD 15 million in initial capacity for AI developers and vendors. It pays on defined AI performance failures, the errors and hallucinations that show up when a model misses a threshold, and it responds without a negligence allegation. It sits apart from cyber and tech errors-and-omissions cover rather than bolting onto them.

Three weeks later, Munich Re’s HSB launched AI Liability Insurance for small and mid-sized businesses. Its pitch is the mirror image of the ISO exclusion: it pays for the AI-related bodily injury, property damage, and advertising injury that general-liability policies now carve out. One Munich Re company is selling the coverage the standard forms just removed. The two moves are the same repricing seen from either side.

What underwriters will ask you to evidence

The specialty products underwrite on things you can measure. aiSure attaches to a model’s defined performance thresholds, so before a policy binds, someone has to state what “performing” means and prove the model stays inside it. If you have taken production readiness seriously, that evidence already exists. It is what the AI Production-Readiness Bar asks for: what the model is allowed to do, how you know it is within bounds, who approves a change. Underwriting is becoming another reader of that evidence, next to your customers’ procurement teams.

You bought the AI, and you still own the risk. What changed in early 2026 is that the risk-transfer market put a price on that ownership, and the price of a clean policy is governance you can show an underwriter.

Building that evidence — the thresholds, the oversight, an owner who can answer an insurer’s questionnaire without a fire drill — is what fractional platform leadership is for: senior review where the risk concentrates, without a full-time hire.