Business Insurance AI Exclusions: Spot CG 40 47, D&O Gaps, and Sublimits at Renewal

A composite scenario assembled from documented cases: In February 2026, the office manager at a 36-person Somerville design consultancy opened the firm's commercial renewal PDF. She checked the two-page declarations sheet and saw a 4% rate bump on a $2 million general liability policy, a $2 million D&O/E&O package, and a $250,000 cyber rider. Relieved, the partners signed that afternoon without reading page 49 of the forms schedule. Three changes hid there: Verisk / ISO form CG 40 47 cutting generative-AI general liability cover (IIABA, Oct 21, 2025), an absolute AI exclusion on D&O and E&O (Jones Day, Apr 2026), and a $15,000 social-engineering sublimit. Six months later, an AI-edited ad drew a trade-dress demand and a deepfake vendor call drained $84,000. Headline limits paid almost nothing.

Skipping the endorsement schedule turns software errors into uninsured cash losses. State regulators approved over 80% of tracked carrier AI-exclusion filings entering 2026 (MoneyGeek, 2026). Even $5 million tech E&O policies can cap chatbot errors at $25,000 (Armilla AI), while Arup lost HK$200 million (~$25.6 million USD) in a February 2024 Hong Kong deepfake video scam (CNN, Feb 4, 2024) and Anthropic's $1.5 billion Bartz settlement across 482,460 books won final approval in July 2026 (Writer Beware, Jul 23, 2026). Paired with our dated casebook of AI mistakes that cost real money, this guide is general operational information, not legal or brokerage advice.

Why did insurers move from silent AI to explicit exclusions in January 2026?

Commercial insurers moved from "silent AI" to explicit exclusions in January 2026 because pre-2023 policy forms never priced generative hallucinations, copyright claims, or automated advice. Standardized exclusions push AI risk off general policies and onto specialty lines where underwriters can price or cap it.

Think of coastal property insurance after severe hurricane seasons. Homeowners policies still pay for a kitchen fire, yet carriers attach a windstorm or flood endorsement removing storm surge unless you buy a separate flood rider. Insurers ran that exact play on cyber risk a decade ago, and in January 2026 Verisk's Insurance Services Office (ISO) applied it to generative AI (IIABA).

Where the flood analogy breaks is traceability. A flood zone sits on a public map; generative AI leaves no physical boundary. ISO defines generative AI across its 2026 forms as "a machine-based learning system or model that is trained on data with the ability to create content or responses, including but not limited to text, images, audio, video or code" (IIABA). Because the forms exclude claims "arising out of" generative AI, an insurer need not prove AI was the sole cause. If an AI prompt contributed anywhere in an assembly manual, ad graphic, or client report, the adjuster can reserve rights or deny defense.

What do ISO forms CG 40 47, CG 40 48, and CG 35 08 cut from general liability?

Verisk / ISO endorsements CG 40 47, CG 40 48, and CG 35 08 (available since January 2026 for Commercial General Liability policies) strip Coverage A physical harm, Coverage B personal and advertising injury, or completed-operations claims tied to generative AI (IIABA). They appear as short alphanumeric codes on your forms schedule rather than page-one alerts.

Standard CGL policies split protection into three buckets. Coverage A pays for third-party bodily injury and property damage. Coverage B pays for listed business torts, including defamation, privacy violations, and infringing copyright, trade dress, or slogans in your advertisements. Products-Completed Operations covers physical harm occurring after your product ships or your job finishes.

Watch CG 40 48 if your team publishes AI commercial images. CGL Coverage B has long been the main defense wallet when marketing triggers copyright, slogan, or likeness claims. User-side AI copyright liability stays unsettled in US courts in October 2026, yet defending even a weak complaint runs $50,000 to $150,000. With CG 40 47 or CG 40 48 attached, CGL pays $0 toward defense.

How do W. R. Berkley's absolute AI exclusions gut D&O, E&O, EPL, and fiduciary policies?

W. R. Berkley's Artificial Intelligence Exclusion (Absolute) (such as form PC 51380) strips AI-related defense and indemnity from Directors and Officers (D&O), Errors and Omissions (E&O), Employment Practices Liability (EPL), and Fiduciary Liability policies (Jones Day). For service firms, SaaS teams, clinics, and nonprofits, these specialty policies are the primary shield for leadership and operations.

Unlike ISO's CGL forms, specialty absolute exclusions often define AI broadly enough to capture predictive scorers, ranking algorithms, and embedded SaaS tools. Their "based upon, arising out of, or attributable to" preface hits four core exposures:

1. Client Deliverables and Chatbots (E&O): Bars negligence claims over hallucinated citations, broken formulas, or invented chatbot promises. Under Moffatt v. Air Canada (2024 BCCRT 149), firms own negligent statements made by website bots; pair policy checks with our chatbot promises liability test and escalation script.
2. AI-Washing Claims (D&O): Bars investor, lender, or FTC/SEC actions alleging executives overstated AI automation or accuracy.
3. Hiring and HR Bias (EPL): Bars discrimination or statutory notice suits tied to AI resume screeners or interview scorers. As cases like Mobley v. Workday advance, an EPL AI exclusion removes defense right when HR needs counsel; use our 12-question AI hiring vendor risk checklist.
4. Governance and Benefit Plans (D&O & Fiduciary): Bars claims alleging inadequate internal AI training or oversight, as well as fiduciary suits when third-party administrators use algorithms to deny health or disability claims. Nonprofit boards carry the exact same exposure on volunteer D&O renewals.

Why does a $250,000 cyber policy pay only $10,000 to $25,000 on a deepfake wire scam?

A $250,000 cyber policy often pays just $10,000 to $25,000 on a deepfake voice or video wire scam because carriers treat staff-approved wires as "social engineering" or "voluntary parting" rather than direct hacking. Full Funds Transfer Fraud limits apply only when an intruder breaches your bank or network directly without tricking an employee into sending the money.

Worse, most social-engineering riders include a callback condition precedent: the $10,000 to $25,000 sublimit pays only if staff verified the payment change beforehand via a pre-existing phone number on file. Calling the number printed on a spoofed PDF can reduce the payout to $0. Enforce our callback rule script for payment and password requests on every finance desk. Separate enterprise data underscores the stakes: IBM's 2026 Cost of a Data Breach study (n = 602 breached organizations, vendor-sponsored Ponemon research) found unapproved AI tools in 43% of security incidents and an average cost of $6.0 million for AI-enabled malicious breaches (IBM, Jul 29, 2026).

Worked loss math: A 45-person firm before vs after negotiation

Assume a 45-person Massachusetts firm carries $2 million CGL ($2,500 deductible), $2 million E&O/D&O ($10,000 retention), and $250,000 Cyber ($5,000 retention). Compare an unaudited renewal (CG 40 47 attached, $25,000 E&O AI cap, $15,000 deceptive-wire cap) against a negotiated renewal (CG 40 48 removed, internal AI carved into full E&O, $100,000 wire cap with logged callbacks):

• Deepfake supplier wire ($88,000): Unaudited pays $15,000 ($73,000 cash loss, or $88,000 without callback). Negotiated (SIM-swapped vendor line despite callback) pays $83,000.
• AI ad trade-dress suit ($64,000): Unaudited CGL with CG 40 47 pays $0. Negotiated CGL pays $61,500.
• AI-assisted report error ($110,000): Unaudited E&O with $25,000 sublimit pays $25,000 ($85,000 loss). Negotiated E&O pays $100,000.

One 45-day renewal check saves $59,000 to $85,000 per claim, usually for $800 to $3,500 in annual premium.

Which carriers offer affirmative AI coverage, and what can you negotiate?

Specialty underwriters including Coalition, Armilla AI, and Mosaic Insurance (with Munich Re's aiSure) now sell affirmative AI coverage that explicitly prices AI security events, chatbot hallucinations, or model performance errors. Most 10-to-250 person firms can also close major gaps on standard renewals by asking brokers for four wording edits.

Ask your broker for four edits before binding: (1) carve out internal office AI (such as M365 Copilot or Google Workspace used with human review) from D&O/E&O exclusions; (2) narrow "arising out of" to "solely and directly caused by"; (3) preserve D&O/EPL defense costs through final adjudication; and (4) raise deceptive-wire caps to $100,000-$250,000 backed by dual-person wire sign-off.

How to audit your renewal in 20 minutes: The ed.cafe 10-Line Audit & Broker Email

Open your specimen policy PDF 45 days pre-renewal and Ctrl+F seven terms: 40 47, 40 48, 35 08, Artificial Intelligence, Generative, Machine, and Social Engineering. If CG 40 48 remains on your CGL, run a four-step pre-publish AI image check: ban living-artist or brand prompts, use enterprise tools with IP indemnity, run similarity checks, and log human edits. Add broker notice contacts to your first-hour AI incident response plan so late notice never forfeits claims-made cover.

Use the ed.cafe 10-Line Insurance Renewal AI Endorsement Audit & Broker Email below on every annual renewal. The ed.cafe 10-Line Insurance Renewal AI Endorsement Audit puts the PDF scan and broker script in one ungated block.

THE ED.CAFE 10-LINE INSURANCE RENEWAL AI ENDORSEMENT AUDIT & BROKER EMAIL
Publisher: ed.cafe by No Worker Left Behind (Boston 501(c)(3), EIN 88-3268312)
Run 45-60 days pre-renewal across CGL, BOP, D&O, E&O, EPL, Fiduciary & Cyber.
Operational checklist; review exact policy wording with a licensed broker.

PART 1: 10-LINE POLICY PDF SEARCH CHECKLIST ("Schedule of Forms & Endorsements")
[ ] 1. CGL Form CG 40 47 attached? (Cuts Coverage A physical + Coverage B ad injury)
[ ] 2. CGL Form CG 40 48 attached? (Cuts Coverage B ad/copyright/likeness injury)
[ ] 3. CGL Form CG 35 08 attached? (Cuts Products/Completed Ops for GenAI designs/code)
[ ] 4. Manuscript AI exclusions? (Search PDF for "Artificial Intelligence" & "Generative")
[ ] 5. D&O / EPL / Fiduciary "Absolute" AI exclusion? (e.g., Berkley PC 51380)
[ ] 6. Tech E&O hidden AI sublimit? (Watch for $10k-$25k chatbot/model error caps)
[ ] 7. Causation breadth: "arising out of / attributable to" vs "solely caused by"?
[ ] 8. Cyber Social-Engineering / Deceptive Transfer sublimit < $100k? (Watch $10k-$25k)
[ ] 9. Crime/Cyber callback condition precedent matches staff phone-verification SOP?
[ ] 10. Claims-made notice window wired into internal AI Incident Response Plan?

PART 2: COPY-PASTE BROKER EMAIL (SEND 45 DAYS BEFORE RENEWAL)
Subject: [Org Name] Renewal - Written Confirmation on AI Exclusions & Sublimits

Hi [Broker Name],

Ahead of our [Renewal Date] renewal across CGL/BOP, D&O, E&O, EPL, Fiduciary, and
Cyber/Crime, please confirm these 6 points in writing with our quotes:

1. CGL Forms: Do any quotes attach ISO CG 40 47, CG 40 48, CG 35 08, or carrier AI
   exclusions? If so, please quote an alternative carrier or request removal of CG 40 48
   so human-reviewed marketing stays covered under Coverage B.
2. D&O / E&O / EPL / Fiduciary: Do any quotes attach an "Artificial Intelligence
   Exclusion (Absolute)" (e.g., Berkley PC 51380) or an AI sublimit below full limits?
3. Internal AI Carve-Back: Where AI wording is mandatory, please request a carve-back
   preserving full cover for internal enterprise productivity AI used with human review.
4. Causation & Defense Costs: Please request narrowing "arising out of" to "solely and
   directly caused by," plus D&O/EPL defense-cost cover through final adjudication.
5. Deepfake Wire Sublimits: Confirm our Social Engineering / Deceptive Transfer cap,
   exact callback prerequisites, and pricing to quote a $100,000-$250,000 sublimit.
6. Affirmative AI Cover: Given our workflows, please advise if Coalition, Mosaic, or
   Armilla AI endorsements are appropriate and share comparative quotes.

Please include the complete Schedule of Forms and specimen copies of any new exclusions.

Thank you,
[Your Name & Title]

Strong verification habits make any team easier to insure. Massachusetts residents can take the Google AI Professional Certificate free via the Mass AI Hub, and No Worker Left Behind is forming Circle #1 at ed.cafe/start/, a free 30-chair cohort closing December 1, 2026 (pre-launch as of October 2026; zero cohorts completed yet).

Where this stops being true

• State filings and surplus lines: US insurance rules vary by state. State regulators approved over 80% of tracked AI-exclusion filings (MoneyGeek), while surplus-lines (E&S) carriers operate outside standard state filings and attach broader manuscript exclusions.
• Commercial AI builders: Incidental-use carve-backs do not protect vendors selling foundation models, clinical AI, or automated underwriting; buy dedicated tech E&O and affirmative AI liability cover.
• Unsettled AI copyright rules: Removing CG 40 48 or holding vendor IP indemnity never covers intentional copying while US output-similarity law remains unsettled.
• Not brokerage or legal advice: Review specimen policies with a licensed broker or coverage counsel before binding or reporting a claim.

Questions people ask

Key terms

• Affirmative AI coverage: Policy wording or standalone insurance that explicitly grants protection for specified AI security events, model errors, or liabilities.
• Arising out of: Broad causal-chain exclusion wording allowing insurers to deny claims whenever AI played any originating or contributing role.
• Coverage B (Personal and Advertising Injury): General-liability section covering business torts such as defamation, privacy invasion, and advertising copyright or slogan infringement.
• ISO CG 40 47 / CG 40 48 / CG 35 08: Verisk commercial general liability endorsements available since January 2026 that exclude generative-AI exposures.
• Silent AI: Unpriced coverage ambiguity in older policies that neither explicitly grant nor explicitly exclude AI-related claims.
• Social-engineering sublimit: Low payout cap (often $10,000 to $25,000) inside cyber or crime policies when staff are tricked into voluntarily sending funds.

Sources

1. IIABA, "Verisk General Liability Exclusions for Generative AI (CG 40 47, CG 40 48, CG 35 08)" (Oct 21, 2025): IIABA
2. MoneyGeek, "Generative AI Exclusions in General Liability (>80% Filing Study)" (2026): MoneyGeek
3. Jones Day, "Maximizing Insurance for AI Risks Amid Emerging Exclusions" (Apr 2026): Jones Day
4. Armilla AI, "Cover for Losses Caused by AI Chatbot Errors" (2025): Armilla AI
5. CNN, "Deepfake CFO Scam Costs Firm HK$200M" (Feb 4, 2024): CNN
6. Writer Beware, "Anthropic $1.5B Settlement Approved" (Jul 23, 2026): Writer Beware
7. Coalition, "Active Cyber Insurance & AI Risk": Coalition
8. Mosaic Insurance, "Munich Re aiSure Partnership": Mosaic Insurance
9. BCCRT, Moffatt v. Air Canada, 2024 BCCRT 149 (Feb 14, 2024): BCCRT
10. IBM, "2026 Cost of a Data Breach (n = 602)" (Jul 29, 2026): IBM