Insurers Step Into AI Risk
Insurers are beginning to offer coverage for business errors caused by autonomous AI “agents,” with some firms cautiously underwriting AI‑related blunders while others hold back reported. That shift creates both a sales aid and a new objection — customers want proof your automation is auditable and insurable.
Lloyd’s coverholders and underwriters began selling affirmative AI‑failure products in 2025, with Armilla announcing an AI liability policy underwritten by Chaucer at Lloyd’s in April 2025. armilla.ai Armilla’s public underwriting playbook shows pre‑deployment model assessments and performance triggers — the firm said a policy might pay out if a model’s accuracy slips from 95% to 85% — and brokers warn many legacy tech E&O policies still impose low AI sublimits (examples cited: $25,000 AI sublimit inside a $5M policy). theoutpost.ai Regulatory pressure is raising auditability standards: the NAIC’s Model Bulletin and related state guidance require documented AI Systems (AIS) programs and oversight for insurers using models, and Lloyd’s Market Association survey (mid‑2025) rated professional‑indemnity losses from AI as a “plausible” scenario while 67% of underwriters expected AI use to rise in the next 12 months. kennedyslaw.com Market friction is visible in commercial real estate: brokers reported a CRE firm that could not treat an autonomous AI agent as an employee and instead needed a bespoke policy, while some carriers are inserting “absolute AI exclusion” clauses as others extend explicit “AI malfunction” language into professional‑services cover. economictimes.indiatimes.com New supply‑side platforms and startups are emerging to standardize underwriting and audits — Testudo launched a Lloyd’s‑backed data platform to price AI litigation risk and the Artificial Intelligence Underwriting Company (AIUC) debuted with a $15M seed and a proprietary audit/standard (AIUC‑1) to tie coverage to audit outcomes. testudo.co Analysts see a growing premium pool: Deloitte’s modeling projects roughly $4.7–$4.8 billion in annual AI‑specific premiums by 2032, and mainstream reporting notes insurers are using underwriting requirements as levers to force guardrails (third‑party audits, versioned logs, SLAs and documented monitoring) before covering agentic AI. deloitte.com
