AI Liability · Standalone · Eight insuring agreements
AI Liability
Standalone cover for the AI your organization actually uses: a person working with a model, an automated decision system, and an agent that acts on its own.
Agreements 1 to 5 · First party · Pays on discovery. No claim required.
Your own loss
Five agreements that respond to your own loss the moment a responsible officer discovers it. No claimant, no lawsuit, no duty to defend. Each measures a different consequence of the same act: money moved, data destroyed, resources burned, the response you had to run, and the decisions you had to redo.
IA 1
Autonomous Execution Loss
Direct financial loss when an agent executes a transaction or a binding commitment incorrectly. It reaches both acting beyond delegated authority and performing an authorized task wrongly.
Triggers when
You discover the loss during the policy period, from an AI event on or after the retroactive date. Recoveries, avoided costs and retained value are deducted.
Where your tower stops
Crime needs a dishonest person and cyber needs a security failure. The agent was authorized and the instruction was yours.
IA 2
Model and Data Restoration
Reasonable cost to restore or reconstruct data and system configurations the AI damaged, deleted or corrupted, back to substantially their prior state.
Triggers when
Discovery, as for IA 1. Customer data is included only within the scheduled restoration scope, and retraining, replacement and betterment sit outside the grant unless endorsed.
Where your tower stops
Cyber restoration needs a security failure. Data is not tangible property, and vendors cap at fees paid.
IA 3
Resource Overrun
Unintended excess compute, API and other scheduled metered charges, including accidental loops and repeated tool calls.
Triggers when
Charges above the reasonable charge for the intended task, net of supplier credits and refunds. Sublimited by default.
Where your tower stops
Nothing answers. The bill is a real metered charge from your own vendor, and planned consumption was never the exposure.
IA 4
Event Response Costs
External cost to investigate and contain a specific AI event, take legal advice on notification duties, and give the notifications required.
Triggers when
An event affecting your own deployment. No third-party claim is required. Prior consent applies, with an emergency containment allowance.
Where your tower stops
A liability form defends a claim, and there is no claim yet. Cyber response needs a security incident.
IA 5
Wrongful Decision Remediation
Cost to identify, review and correct affected decisions when a scheduled automated decision process produces materially erroneous or unlawfully discriminatory outcomes.
Triggers when
Discovery of the obligation, for an identified affected population. Corrective work only, not compensation owed to claimants.
Where your tower stops
Nothing does. Liability forms pay the individual claimant, never the cost of redoing the population.
Agreements 6 and 7 · Third party and regulatory · Claims made and reported.
Third-party and regulatory claims
Two agreements that answer to someone outside the business: a claim for wrongful disclosure through output, and the cost of defending a regulatory proceeding within your elected scope. Defense costs erode the applicable sublimit and the aggregate unless an endorsement says otherwise.
IA 6
Wrongful Output Disclosure
Damages and defense costs for a claim alleging wrongful disclosure of protected information caused by an AI event in covered use.
Triggers when
A claim first made during the policy period. A network intrusion or security failure is expressly not required.
Where your tower stops
Cyber privacy needs unauthorized access or a security failure. An over-permissioned system breached nothing.
IA 7
AI Regulatory Proceedings
Reasonable defense and investigation costs in a regulatory proceeding within the scope elected in your schedule.
Triggers when
A formal investigation or proceeding first commenced during the period. Fines, penalties, taxes and mandated customer redress are not included.
Where your tower stops
A regulatory proceeding is not a claim for damages, and a general market inquiry is not a proceeding.
Agreement 8 · Elected head by head · Claims made, each head separately elected.
AI Liability, elected head by head
Agreement 8 carries five liability heads. Each is separately elected, separately scoped and separately attached: primary where you schedule and price it for the exposure, or difference in conditions against an identified gap in a named underlying policy. No head is included because another was purchased, and an absent, denied or exhausted underlying policy does not by itself create attachment.
Head 8A
Professional and Operational Error
Third-party financial harm from erroneous AI output or execution in your scheduled business activities, including customer negligence claims.
Triggers when
A claim where covered AI use materially contributed to the alleged wrong. Elected on a primary or difference in conditions basis.
Where your tower stops
Professional liability, but only with a client claimant, the insured activity, and no AI exclusion.
Head 8B
Automated Decision Liability
Discrimination and other elected employment or consequential decision claims affecting employees, applicants or customers.
Triggers when
A claim. Claimant damages here are separate from the IA 5 cost of correcting the decisions themselves.
Where your tower stops
Employment practices reaches the employee only, never the consumer, tenant or credit applicant.
Head 8C
Content and Publication
Defamation and expressly elected intellectual property infringement arising from AI assisted content.
Triggers when
A claim from what you published. Patent and trade secret claims are not automatically included, and require explicit election.
Where your tower stops
General liability advertising injury now carries the ISO exclusions. A vendor indemnity moves money, not the duty.
Planned
Head 8D
Bodily Injury and Property Damage
Physical harm from expressly endorsed activities only, with a scheduled operating envelope and its own sublimit.
Triggers when
A claim under the endorsed activity. This is not automatic safety critical, clinical or autonomous vehicle cover, and it requires specialist review.
Where your tower stops
General liability, now the widest US gap after the ISO exclusions took effect 1 January 2026.
Planned
Head 8E
Management and Fiduciary
Claims concerning scheduled AI governance duties or representations, in specified insured capacities.
Triggers when
A claim against a person in that capacity, with D&O and fiduciary coordination and securities treatment named in the endorsement.
Where your tower stops
D&O and fiduciary, where no AI exclusion is attached. Filed forms now exclude all three together.
Agreements marked Planned are not automatic launch cover: each needs an express endorsement and specialist review. What is shown describes the intended cover, and the policy wording governs in every respect.
Who buys this
Illustrative buyers, and what they take
Twelve representative placements, hypothetical rather than customers. Every agreement is optional, and some buyers take Agreements 1 to 7 as a package. Indicative premiums are demonstration outputs of the Rating Plan, not quotations, and exclude the policy fee. Every figure is subject to wording, elections, verified exposure and an approved ratebook.
Buyer
Uses AI to
Agreements
Indicative premium
Rating driver
Wholesale distributor
Purchasing agents, ERP copilots and customer service tools.
Whole policyHead 8AHead 8BHead 8C
USD 62,000 to 78,000
USD 40M authorized purchasing
Payments business
A treasury agent and internal reconciliation automation.
IA 1IA 2IA 4
USD 15,000 to 19,000
USD 20M authorized payments
Freight brokerage
Booking agents and dispatch automation.
IA 1IA 4Head 8A
USD 24,000 to 30,000
USD 12M booking commitments
Professional services group
Research, billing, client support and recruitment workflows.
Whole policyHead 8AHead 8BHead 8C
USD 78,000 to 96,000
300,000 sensitive client records
Marketing agency
Creative copilots and an external campaign agent.
IA 2IA 3IA 4
USD 10,000 minimum
USD 600,000 metered spend
Employer
A licensed screening tool and an internal scheduling agent.
IA 4IA 5IA 7Head 8B
USD 55,000 to 70,000
25,000 consequential decisions
Manufacturer
Engineering assistants and a records scheduling agent.
IA 2IA 4
USD 10,000 to 13,000
180,000 writable planning records
Retail group
Refund, stock, service and HR workflows.
Whole policyHead 8AHead 8BHead 8C
USD 68,000 to 84,000
USD 30M refund authority
Property manager
Leasing assistants, maintenance automation and a public chatbot.
IA 4IA 6Head 8C
USD 26,000 to 33,000
120,000 tenant records
Logistics operator
Scheduling agents and an AI controlled handling system.
IA 1IA 4Head 8D
USD 90,000 to 120,000
Endorsed handling activity
Listed company
Internal AI tools, with governance representations to investors.
IA 4Head 8E
USD 60,000 to 85,000
Scheduled insured capacities
Outsourced services firm
Document agents and billing automation in its own service delivery.
IA 1IA 2IA 3IA 4IA 5IA 6Head 8A
USD 52,000 to 66,000
350,000 client records
Shaded agreements are available at launch. "Whole policy" marks buyers who take Agreements 1 to 7 as a package, with the heads shown elected on top.
Underwriting
Application and a public-record check
An application and exposure schedule, plus a check of public sources. No audit, no penetration test, no model evaluation, and nothing installed in your environment. Evidence is requested only where a material fact is unresolved and resolving it would change the decision.
Through your broker
Placed through appointed surplus lines brokers. Your broker brings the application; there is no self serve gate and no pre bind audit.
Declare, then schedule what matters
You declare your AI use, then schedule the workflows that can move money, change records, disclose or publish, make consequential decisions, spend metered resources or control equipment. Low consequence use is declared by category, not scheduled one tool at a time.
Evidence only where it changes the answer
Where a permission, cap, approval or stopping control is unclear and the answer would change the decision, you get one consolidated request naming the workflow, the fact needed and the decision being held. We do not ask for every available log.
A fast, plain answer
An indication within five business days of a complete submission, with the reason and what is still outstanding. A hard decline comes within one business day, with the reason.
Auxilium Governance · included free
Included with every policy. Never a condition of coverage.
We ask which governance tooling you use, and we do not require any particular one. Connecting Auxilium Governance is not a condition of cover and never determines whether a claim is paid. What it does is make facts about your AI use cheap to establish, which is what shortens underwriting.
Not a condition, and not a warranty
Software enrollment is never an automatic condition of coverage, and an application answer does not by itself become a warranty. Any control that is a condition must be specific, measurable and expressly endorsed.
Any tooling qualifies
What matters is the capability rather than the vendor: an inventory of AI use, visible permissions, attribution to an actor, and a retained window of activity. Ours, another provider's, or one you built, all count the same.
It buys a better decision, not a discount
Continuous telemetry narrows the uncertainty around what your agents can do and how fast they can do it. Narrower uncertainty means fewer referrals and a faster answer, and that benefit is real from the first submission.
