Restoration Intel Brief — September 26, 2026
What happened
On September 24, specialty insurer Beazley launched two cyber endorsements for the risks of running your own AI — not the risk of AI attacking you, the risk of your AI going wrong:
- AI Voluntary Shutdown (described as an industry first): if you have to suspend your own malfunctioning AI system before it causes significant financial or reputational harm, this covers the business-interruption and lost-earnings cost of that precautionary shutdown.
- AI Regulatory Defence & Penalties: if unintentional misuse of your AI triggers a regulatory investigation or fine, this covers defense costs and penalties under AI-specific legislation.
Source: Beazley announcement · Insurance Business coverage
Why this matters for contractors running AI
This is the first-party half of the coverage gap. The other half is third-party liability — who pays when your AI harms someone else (GL exclusions, tech E&O). This is about your costs when your own system misbehaves:
- Your AI voice line takes a live homeowner call at 2am and starts booking jobs wrong, quoting wrong, or saying things it shouldn’t. You pull the plug at 3am. The shutdown endorsement is built for exactly that morning-after cost.
- Your agent harness drafts and sends customer emails with a hallucinated price. You shut the lane down to fix it. Same thing.
- A regulator comes asking about how your AI handles customer data or makes decisions. The regulatory endorsement answers the investigation and the fine.
Beazley’s cyber chief, Alessandro Lezzi, framed it as a duty-of-care point: insurers should state explicitly that AI incidents like these are covered rather than leaving it ambiguous. He also said he expects this kind of product to become close to standard across the market over time.
The context worth knowing
In April, the Financial Times reported Beazley was among insurers proposing policy language that would cap payouts for certain AI-linked losses, including regulatory breaches. Brokers pushed back. In early September Beazley confirmed affirmative AI wording (AI-driven attacks covered under existing cyber/Tech E&O). Now these two endorsements. Read in sequence, that’s a carrier moving toward more explicit AI cover, not less — but the April episode is why you read the actual wording instead of the press release.
What’s still unclear (broker questions)
- Is this wording available to US buyers through Beazley’s USA platform, or London-market only — and on admitted or surplus-lines paper in Washington?
- Does “voluntary shutdown” require insurer pre-approval, or is notifying after the fact enough? Waiting periods? Sublimits?
- How does the wording define “malfunctioning” — does a precautionary shutdown based on suspected (not proven) misbehavior trigger cover?
- On the regulatory side: does it respond to state attorney-general actions and consumer-protection investigations, or only breaches of AI-specific statutes?
- How does it interact with the ISO generative-AI exclusions (CG 40 47 / 40 48 / 35 08) carriers are attaching to GL at renewal — is there a gap between the GL exclusion and where this endorsement picks up?
Suggested next step
Forward this to your broker alongside your AI coverage questions. The one-line ask: “We’re running AI voice and agent systems in production — quote us the first-party shutdown and regulatory-defense wording, and show us where it overlaps or gaps against our current cyber and E&O.”
Not legal advice. Insurance is state- and carrier-specific; the market is mid-rewrite in 2026. Verify wording with your broker before relying on it.
Restoration Intel tracks carrier moves, claims practice, and AI risk for restoration operators. More briefs →