In early 2024, an employee at the UK based engineering firm Arup joined a video call with people who looked and sounded like the company’s CFO and colleagues. Every face on the call was a deepfake/synthetic. By the end of it, roughly $25 million had been wired to the attackers.

Now hold that number against a fact most finance teams don’t know: the social engineering fraud sublimit on a typical cyber policy sits between $100,000 and $250,000 (IRMI). That is a 100-to-1 gap between what a real deepfake attack can cost and what your insurer has quietly agreed to pay.

Here is the uncomfortable part. Whether your policy responds to a deepfake loss at all is not decided by the size of your limit, your premium, or the reassuring word “cyber” on the cover page. It is decided by definitions, triggers, sub-limits and conditions buried in the wording and in 2026, 2027 that wording is moving against policyholders by default.

This is the one idea worth internalising before your next renewal: a deepfake claim is won or lost in the language, and the language changed on 1 January 2026. Let’s walk the ten traps in the order a claim actually collapses.

One question I keep hearing is: “Would my cyber insurance actually cover a deepfake fraud?” To make that easier to investigate, I built a simple Deepfake Insurance Coverage Checker. You can answer a few questions about a potential fraud scenario or paste relevant sections of your cyber/crime policy wording. The tool highlights important terms and phrases and explains what they could mean in plain English, particularly around issues such as impersonation, social engineering, funds-transfer fraud, and deepfake-enabled payment instructions.

It’s designed as a quick first-pass research tool for finance, IT, risk, compliance, and insurance teams, not a substitute for your broker, insurer, or legal advice. One useful feature: what you enter stays in your browser.

Try the Deepfake Insurance Coverage Checker: Deepfake Insurance Coverage Checker

Few snapshots of the checker tool:

Most deepfake losses never reach an exclusion. They fail earlier, because the insuring agreement was never built to catch them.

Trap 1: The “voluntary parting” exclusion
When an employee is deceived into authorising a transfer, no firewall was breached and no system was compromised - a human clicked “approve.” Insurers treat that as the insured voluntarily parting with funds, and a standard exclusion removes coverage even when the instruction was fraudulent. As a Jones Walker analysis put it, the exclusion bites because your agent willingly parted with the money. The quality of the deepfake is, under standard wording, legally irrelevant. This is the single clause most likely to sink your claim.

Trap 2: The synthetic-media definition gap
Many social engineering endorsements were written for human deception “impersonation,” “pretexting,” “fraudulent instruction.” They never name AI-generated audio, video or text. If the definition doesn’t explicitly reach synthetic media, an insurer can argue a deepfake event simply isn’t the thing you bought cover for. Since 1 January 2026 the market has openly split: some carriers now affirmatively cover AI-generated impersonation (BOXX added an explicit AI/deepfake endorsement to its Cyberboxx Business policy in mid-2026), while others excluded it from standard social engineering wording. You cannot assume which side your carrier landed on.

Trap 3: The sublimit masquerading as coverage
Even with a clean trigger, social engineering and funds-transfer fraud almost always carry a dedicated sublimit far below your headline policy limit. A $5M cyber policy that caps social engineering at $250,000 is a $250,000 policy for this risk. The best endorsement language in the world pays nothing above the sublimit, so the sublimit, not the aggregate limit, is the number that decides whether your coverage is real or performative.

Trap 4: Trigger mismatch and the “direct loss” fight
Policies slice deception into distinct buckets: social engineering fraud, funds transfer fraud, and computer fraud. Some cover one and not the others. In a worse case, these agreements typically require a “direct” loss, and courts are split on whether a loss routed through a deceived human is “direct” enough to trigger. As Wiley firm notes, “direct loss” has become a pivotal battleground in social-engineering coverage, with AI-generated impersonation now at its centre. A deepfake inserts an extra human decision into the chain, precisely the link insurers use to argue the loss is indirect.

Even if the situation qualifies as a covered deepfake incident, two other conditions in the policy can still prevent the insurer from paying the claim.

Trap 5: Strict verification or callback requirements
Some insurance policies say that a payment will only be covered if the company follows specific verification procedures, for example, calling a trusted phone number or getting approval from a second person.

These requirements made sense when the policies were written, but they can create problems with AI-generated impersonation in 2026.

For example, imagine your employee follows the company's verification process, but the deepfake is convincing enough to defeat it. The employee might call a spoofed phone number or get confirmation from a second person who is also seeing the same fake AI-generated identity.

Even though your team genuinely followed its procedures, the insurer could still argue that the policy's exact verification condition was not satisfied and deny the claim.

London broker New Dawn Risk has advised clients to avoid policies with rigid verification requirements.

If the insurer will not remove the requirement, try to negotiate an exception such as:

Trap 6: Putting the coverage only on a cyber policy
Deepfake payment fraud can look like a cyber incident, but it is often legally and practically treated as crime or fidelity loss.

February 2026 guidance from PLUS and Kennedys makes this distinction clearly: deepfake fraud is fundamentally a crime/fidelity risk rather than simply a cyber risk.

This creates a potential problem if your deepfake protection exists only as an extension of your cyber insurance. An insurer could argue that the loss is really a crime claim, not a cyber claim, creating a coverage mismatch and a dispute over which policy should respond.

For example, if an employee is tricked by a deepfake CEO into sending a wire transfer, the loss may be better covered under a crime or fidelity policy than under cyber insurance.

The practical lesson: For wire-fraud and deepfake-payment risks, don't rely on cyber insurance alone. Make sure the risk is covered under a crime/fidelity policy, or that the relevant coverage is clearly coordinated across both cyber and crime policies.

Even a well-placed, well-triggered claim can hit an exclusion that didn’t exist at your last renewal.

Trap 7: The AI-generated-content exclusion added at renewal
On 1 January 2026, ISO/Verisk introduced generative-AI exclusion endorsements - CG 40 47, CG 40 48, CG 35 08, for commercial general liability, and carriers adopted them within weeks (Fenwick, Shumaker). Those specific forms sit on CGL, not cyber but the signal matters, because several cyber and Tech E&O carriers filed their own proprietary AI exclusions tracking that language. The lesson: whether AI is excluded is a question about your policy’s endorsement schedule, not its renewal date or its line of business. Read the schedule, page by page.

Trap 8: Mistaking a reputational-harm endorsement for fraud cover
In December 2025, Coalition launched a purpose-built Deepfake Response Endorsement, a genuinely useful product for forensic analysis, takedowns and crisis communications after synthetic-media reputational harm. It does not cover wire-transfer losses. Buying it and assuming your $25M transfer scenario is now handled is a category error: it addresses reputation, not funds out the door. Know exactly which risk each endorsement answers.

Trap 9: War, hostile-act and systemic-event exclusions
After the NotPetya disputes, cyber policies became more explicit about war and state-backed cyberattacks. If a deepfake fraud is attributed to a nation-state or state-linked actor, an insurer may try to invoke a war, hostile-act or state-backed-attack exclusion but state involvement alone does not automatically exclude the loss. Coverage depends on the policy’s exact wording, attribution rules and carve-backs. Check who determines attribution, what qualifies as a state-backed attack, and whether an exception preserves coverage.

Trap 10: Controls, notice and consent conditions
Underwriters are increasingly interested in the controls an insured actually has in place, not just what was described during underwriting. PLUS/Kennedys notes that post-incident scrutiny focuses on the procedures that were actually in place.

If an application says the company uses callback verification, dual authorisation or multi-factor approval, but the investigation shows those controls were missing or not followed, the insurer may challenge coverage depending on the policy wording and applicable law. Other procedural requirements can create problems too, including prompt-notice obligations, consent-to-settle provisions and retroactive-date or prior-acts restrictions. A loss can therefore appear to fit the main coverage grant but still become disputed because of a condition or exclusion.

“We have cyber insurance, so a deepfake wire fraud is covered.”

Cyber insurance was built for breaches, ransomware and outages, events where a system was compromised. A deepfake-induced transfer compromises a person, not a system. That’s why the loss keeps landing in the crime/fidelity world, why sublimits are small, and why the voluntary-parting and verification clauses exist. The word “cyber” on the policy is not a coverage grant. The wording is.

The deepfake threat is not primarily a technology gap. It’s a contract gap. Attackers have already crossed the capability line, cloned voices and live video fakes are cheap and convincing, and the reinsurance market (Swiss Re’s SONAR among others) has flagged deepfakes as a driver of future cyber losses. The market’s response in 2026 has been to draw the coverage boundaries sharper, some carriers affirming, more of them excluding, and almost all of them tightening the definitions that decide who pays.

That means the advantage now goes to the buyer who reads. Two SMBs with identical premiums and identical logos on the cover page can have opposite outcomes on the same $2M deepfake wire fraud, decided entirely by ten clauses most people never open.

Actionable Items: do these three things this quarter:

It is a free weekly briefing you can read in under 10 minutes - covering how deepfakes are actually built, real deepfake fraud cases studies, honest detection tool reviews, and prevention protocols that work. Written independently. 1,000+ cybersecurity professionals, founders, and IT managers across SMBs already read it. Feel free to forward it to your friends who might benefit from this newsletter.

Have you seen something that didn't feel right? or Got questions you don't know who to ask? Whether you've encountered a suspected deepfake, want to understand your exposure, or just want to talk through what's happening in this space, I'm setting aside time for free 30-minute calls.