What it covers
- Model error and drift
- Hallucinations and harmful outputs
- Data leakage
- Bias and discrimination claims
- Defence costs
- Insurable regulatory fines, with the EU AI Act and Colorado AI Act named
Watch out for
- The per model sublimit is $5M inside the $25M aggregate
- It sits alongside Tech E&O, it does not replace it; allocation between the two policies is the placement question
- No specimen wording was available in our research; the grants above are from Armilla's published product materials
Who it fits
- Companies building and deploying AI models
- AI vendors whose customers require dedicated AI limits
Who it does not fit
- A SaaS company with no AI product (an endorsed Tech E&O is usually the right buy)
Frequently asked questions
Is Armilla sold direct to startups?
No. Armilla distributes through licensed brokers, which is precisely why a retail broker can offer it and the direct platforms cannot offer a market comparison of it.
What limit can one model get?
Published capacity is $25M aggregate per insured with a $5M per model sublimit, raised in January 2026.
Source: Program details from Armilla's published company and product materials and Lloyd's capacity announcements, as compiled in our AI risk research checked November 14, 2026.
Talk through your stack before you sign the next contract
Send us the term sheet, the customer MSA, or the renewal you are worried about. We will tell you plainly what it requires, what your current programme does about it, and what we would place differently.