Insurance provider Cowbell will launch a standalone policy from next year, covering US small and medium-sized enterprises (SMEs) against losses from uncontrolled artificial intelligence (AI) agents.
The product, titled Cowbell AI Liability Insurance, will be distributed exclusively through wholesale brokers on a non-admitted basis.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
It is tailored for commercial entities utilising AI tools across customer support, payment processing, and primary operational tasks.
Claims are triggered by an “AI Incident”, defined as an occurrence where an AI tool breaches its authorised boundaries, operates without necessary clearance, exposes data without authorisation, produces inaccurate or hazardous output, or evades established protocols.
Cowbell stated that qualifying events also encompass model drift, algorithmic discrimination, unstable behaviour, and adversarial manipulation.
Potential incidents include automated systems incurring unauthorised usage fees through runaway loops, executing discriminatory decisions without human oversight, or being manipulated into disbursing fraudulent payments.
Cowbell co-founder and COO Trent Cooksley said: “The financial consequences can reach across the business, from money lost in a transaction to interrupted operations and customer claims. We designed Cowbell AI Liability Insurance to help businesses manage those risks as they put AI to work.”
Under first-party provisions, the policy may compensate for the costs of isolating impacted tools, investigating events, preserving evidence, recalling or correcting defective outputs, and notifying recipients.
Protection additionally extends to direct financial transaction losses, charges triggered by looping or escalating tasks, and LLMjacking incidents involving stolen API keys or credentials exploited by unauthorised third parties.
For third-party liabilities, coverage may cover defence costs, damages, or settlements connected to algorithmic bias, copyright or trademark infringements, bodily injury, or property damage stemming from autonomous actions or reliance on AI output.
Where permitted by law, protections may also extend to regulatory proceedings, fines, penalties, consumer redress, and punitive damages.
Underwriting assessments will incorporate three Cowbell Factors measuring AI observability, autonomy, and governance, alongside broader underwriting details.
The insurer noted that ultimate coverage remains subject to specific policy terms, conditions, limits, retentions, exclusions, and incident circumstances.
