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.
Seven agreements that respond to your own loss the moment you find it. No claimant, no lawsuit, no duty to defend. One trigger runs through the whole section: a declared agent acted outside its authority, or inside it but against the instruction it was given. Each agreement then measures a different consequence of that same act, from money moved to data destroyed to a process you had to stop.
The financial consequence when an agent departs its authority: what it moved, committed, or lost, net of what you kept or recovered.
Triggers when
A declared agent acts outside its authority, against its instruction, or through manipulation. Departure, not outcome.
Where your tower stops
Crime needs a dishonest person; cyber needs a breach. The agent is authorized and the instruction is yours.
Putting back data, code, weights, or configuration the agent deleted or corrupted.
Triggers when
The same three-limb agent trigger as IA-1, discovered by you.
Where your tower stops
Cyber restoration needs a security failure. Data is not tangible property, and vendors cap at fees paid.
Compute and inference spend a declared agent ran past a limit you set.
Triggers when
Consumption past the limit you configured, or through manipulation. With no limit configured, a reduced recovery applies.
Where your tower stops
Nothing answers. The bill is a real, metered charge from your own vendor.
Finding out what happened and correcting wrong output, before anyone claims.
Triggers when
Your awareness of an event, with our prior consent. Carried on every policy.
Where your tower stops
A liability form defends a claim, and there is no claim yet. Cyber response needs a security incident.
Lost profit and extra expense when a declared vendor fails you, plus migration.
Triggers when
A provider event: a withdrawn or changed model, a ceased service, or a lost license.
Where your tower stops
Cyber contingent interruption needs a breach or outage; a model deprecated on notice is neither.
Lost profit and extra expense while a declared agent is switched off, or while your own systems are down because of what it did.
Triggers when
You suspend the agent after a covered act, a regulator or court requires it, we direct it, or a covered act leaves your systems inoperable.
Where your tower stops
Cyber and property interruption need an unplanned outage. A deliberate suspension is the opposite.
The scheduled cost of re-determining a population of decisions you must redo.
Triggers when
Discovery of the obligation to re-determine, from a statute, regulator, court, or your own policy.
Where your tower stops
Nothing does. Liability forms pay the individual, not the cost of redoing the population.
Agreements marked Planned are not available at launch. What is shown for them describes the intended cover, and the policy wording governs in every respect.
Two third-party agreements, with a duty to defend and defense costs inside the limit.
Damages and defense when your system discloses protected information through its output.
Triggers when
A claim, whether or not any access was unauthorized or the system held valid credentials.
Where your tower stops
Cyber privacy needs unauthorized access or a security failure. An over-permissioned system breached nothing.
Defense, insurable fines, and consumer-complaint response, under a listed AI statute.
Triggers when
A formal investigation or enforcement action under a scheduled statute.
Where your tower stops
A regulatory proceeding is not a claim for damages. Employment practices reaches an employment charge, not this.
IA-10 carries five heads, each separately purchased and attached. A head sits in front of your expiring policy as primary where that policy carries an AI exclusion, and behind it as difference in conditions where it does not. You pay for the gap you actually have.
Third-party loss from wrong output or an autonomous act relied on in your business.
Triggers when
A claim. No requirement that anyone failed to check, or that anything was breached.
Where your tower stops
Professional liability, but only with a client claimant, the insured activity, and no AI exclusion.
Discrimination and employment or consumer wrongs from a system's part in a decision, ML or generative.
Triggers when
Participation in the decision, not authorship. A model that narrows a shortlist has participated.
Where your tower stops
Employment practices reaches the employee only, never the consumer, tenant, or credit applicant.
IP infringement, defamation, and publicity or privacy, from your output or how you trained.
Triggers when
A claim, from what you published or what you did to build the system.
Where your tower stops
GL advertising injury now carries the ISO exclusions. A vendor indemnity moves money, not the duty.
Third-party injury or damage from reliance on output or an autonomous act. On the ordinary submission.
Triggers when
A claim. A clinical claim needs a licensed person's review before the output is relied on.
Where your tower stops
General liability, now the widest US gap after the ISO exclusions took effect 1 January 2026.
Claims against an insured person for AI oversight, disclosures, or benefit-plan administration.
Triggers when
A claim against a person in that capacity, excess of your D&O and fiduciary policies.
Where your tower stops
D&O and fiduciary, where no AI exclusion is attached. Filed forms now exclude all three together.
Hypothetical profiles, not customers. Every agreement is optional, and many buyers take the whole policy. Premiums and estimated exposures are illustrative, within the worked-scenario range on the form.
| Buyer | Uses AI to | Agreements | Premium | Est. exposure |
|---|---|---|---|---|
Law firm | Associates draft and research with models. | IA-10(a)IA-8IA-4 | USD 14,200 | USD 1.2M |
Staffing platform | A model screens and ranks applicants. | IA-10(b)IA-7IA-9 | USD 28,500 | USD 2.5M |
Consumer lender | Automated models decide credit. | Whole policy | USD 34,800 | USD 3.5M |
Health system | Clinical decision support triages and doses. | Whole policy | USD 41,200 | USD 4M |
Logistics operator | An agent books carriers and commits rates. | Whole policy | USD 22,600 | USD 1.8M |
Finance team | An agent reconciles and pays invoices. | IA-1IA-2IA-4 | USD 18,400 | USD 1.5M |
Insurance MGA | An agent triages and adjudicates claims. | Whole policy | USD 24,900 | USD 2M |
Property manager | A model screens prospective tenants. | IA-10(b)IA-9 | USD 12,700 | USD 900K |
Marketing agency | Generative tools produce campaigns. | IA-10(c)IA-4 | USD 11,300 | USD 750K |
Retailer | A pricing agent sets prices in real time. | IA-1IA-10(b)IA-3 | USD 16,800 | USD 1.1M |
Accounting firm | AI drafts filings and reconciles books. | IA-10(a)IA-8 | USD 13,500 | USD 1M |
Engineering firm | AI generates specifications and drawings. | IA-10(d)IA-10(a) | USD 26,300 | USD 2.2M |
Wealth advisor | AI drafts advice and plan allocations. | IA-10(a)IA-10(e)IA-9 | USD 19,700 | USD 1.6M |
Publisher | AI writes and edits published articles. | IA-10(c)IA-8 | USD 15,400 | USD 1M |
Manufacturer | A vendor model runs quality inspection on the line. | IA-10(d)IA-5IA-6 | USD 29,600 | USD 2.8M |
Shaded agreements are available at launch. "Whole policy" marks buyers who take the form as a standalone.
An Application and a public-record check. No audit, no penetration test, no model evaluation, and nothing installed in your environment to price the risk. Four conditions precedent and no others, and failing one affects only the claim it relates to.
Placed through appointed surplus lines brokers. Your broker brings the Application; there is no self-serve gate and no pre-bind audit.
You select the categories of use you want covered and schedule only the systems that can do real damage. For each agent, first-party cover attaches once the three governance controls are evidenced, from your own configuration and logs, a supplier attestation, or an accreditation.
An indication within five business days of a complete submission. A hard decline within one business day, with the reason.
For any agent you run, qualifying governance is a condition of coverage: a documented authorization scope enforced in configuration, timestamped logging, and a person other than the agent who can halt it. Every policy includes Axiom Governance at no cost to put all three in place and keep the evidence, so the AI is governed from the day it is deployed.
Authority the agent cannot exceed, a log of what it did, and a person who can stop it. Nothing else is required, and none of it applies if you do not run agents.
Your own configuration and logs, a written attestation from the supplier of the agent, or an accreditation. The free software is simply the easiest way to produce that evidence and keep it current.
You never pay for it. It reports factual observations only, it never scores or certifies your systems, and staying connected through a policy year earns the retention credit in your Declarations at renewal.
Traditional insurance was not written with any of these in mind. This form is.
CG 40 47, CG 40 48 and CG 35 08, effective 1 January 2026, on the general liability line every commercial buyer holds.
High-risk obligations phasing in, with fines up to EUR 35M or 7% of global turnover.
Automated-decision and AI-transparency laws, tracked on the AI Statute Schedule attached to each policy and maintained quarterly.
AI-disclosure securities actions were pleaded in the first half of 2026, reaching directors and officers cover.
Action against deceptive or harmful AI practices under existing consumer-protection authority.
Professional-responsibility duties for AI use by lawyers.