🎭 Cybersecurity

AI Phishing, Deepfakes, and Voice Cloning: What Small Businesses Must Do in 2026

Attackers no longer need skill or patience. AI writes perfect phishing emails, clones voices in seconds, and runs live video calls with your fake CEO. Here's how the new fraud works and how to stop it.

By EDCON Team · May 24, 2026 · 10 min read

In early 2024, a finance employee at a multinational company joined a video conference call with who appeared to be his company's CFO, several colleagues, and outside advisers. They discussed an urgent wire transfer. He processed it. The total: $25.6 million. Every other participant on that call was an AI-generated deepfake.

That attack required significant resources and was aimed at a large enterprise. By 2026, the same technology runs on a laptop and is being pointed at small businesses in Ventura, Azusa, Oxnard, and every other market where owners can't afford a dedicated security team. This guide explains exactly how these attacks work, what makes them so effective, and the specific defenses that actually stop them.

What changed: fraud before and after AI

Traditional phishing had tells. The grammar was off. The sender's domain was a typo-squat. The urgency felt manufactured. Employees learned to spot these signals over years of security training, and for a while, detection rates improved.

AI erased most of those tells. By mid-2026, 82.6% of analyzed phishing emails show signs of AI generation. In the second quarter alone, 40% of business email compromise (BEC) emails were entirely AI-written. These messages are grammatically perfect. They reference your real vendor names, your actual bank, your correct job title, the project you're currently working on. They arrive during business hours and follow up on real email threads when attackers have compromised an account and can read your inbox.

The financial impact follows from this shift. Average per-incident losses from AI-augmented business email compromise now exceed $4.1 million, compared to $1.3 million for traditional phishing. The attack is harder to detect, the dollar amounts are larger, and the response window is shorter because attackers push for same-day wire transfers.

The three attacks you need to understand right now

1. AI-generated phishing emails

This is the most common attack, and it has changed in a specific way that matters. Older phishing used templates — mass messages with generic salutations and vague urgency. Modern AI phishing is personalized at scale. Attackers use tools that automatically pull data from LinkedIn, your website, social media, and public business records to build a profile of each target, then generate a custom message that feels like it came from someone who knows your business.

For a childcare center owner, that might look like an email from a state licensing agency referencing your actual license number and an "updated compliance form" due in 48 hours. For a small business owner, it might arrive as a message from your accounting software provider explaining a billing issue that requires you to re-enter your bank details through a link that looks exactly like their real login page.

2. AI voice cloning and vishing (voice phishing)

Voice phishing surged 442% between 2023 and 2024. AI is the reason. Attackers can now clone a human voice using as little as three seconds of audio. That audio might come from a YouTube video, a podcast interview, a company webinar, or even a phone voicemail greeting. Commodity tools sold on dark web markets allow voice cloning for under $20, and automated AI systems can now conduct entire conversations using the cloned voice without a human operator on the line.

The attacks typically target employees with payment authority. A call appears to come from the business owner's number. The voice sounds exactly right. The caller explains that a vendor payment needs to be processed urgently while the owner is traveling and can't handle it personally. The employee, hearing a familiar voice and a plausible story, completes the transfer.

Childcare centers face a specific version of this: calls that appear to come from a parent's number, using a voice the staff recognizes, requesting that a child be released to an unfamiliar person or that a payment method be changed. It is a physical safety risk, not just a financial one.

3. Deepfake video calls

This is the attack that makes even experienced security professionals uncomfortable, because it defeats the verification method most people consider the gold standard: seeing a face. Real-time deepfake technology can render a convincing likeness of a known person over live video during a Teams or Zoom call, complete with realistic lip-sync and expressions.

Deepfake-enabled fraud incidents grew 680% year-over-year between 2024 and 2025. The FBI's Internet Crime Complaint Center flagged deepfake-assisted fraud as the fastest-growing category of AI cybercrime in the United States. Generative AI-enabled fraud losses are projected to reach $40 billion by 2027.

The most common scenario targeting small businesses is the fake executive video call: an employee receives an urgent video call request from someone appearing to be the owner or a senior manager. The "executive" explains a confidential acquisition, legal matter, or financial emergency that requires an immediate wire transfer before close of business. They ask the employee to keep it quiet. The employee complies.

Why small businesses are the preferred target

Small and mid-sized businesses account for over 70% of all data breaches. The reason is straightforward: large enterprises invest heavily in security tools, legal teams, and dedicated fraud analysts. Small businesses typically have one person who handles everything from payroll to IT support, and that person is already stretched thin.

Attackers also know that small business owners tend to have direct payment authority without additional approval layers. At a corporation, a $50,000 wire requires multiple sign-offs. At a small business, it might require only the owner's say-so or, worse, a trusted employee acting on what appears to be the owner's instruction.

The AI tools make targeting economical. In 2021, running a personalized phishing campaign against 500 small businesses would have required significant human labor. Today, an attacker can configure an AI to profile targets, draft personalized messages, and send them automatically at scale, then triage responses without manual effort. The cost of attacking each individual target approaches zero.

Six specific defenses that work against AI fraud

1. Establish a verbal verification policy for all financial requests

This is the single most effective countermeasure and costs nothing to implement. Any request involving a wire transfer, payroll change, vendor payment modification, or account credential change must be verified through a separate phone call using a number you already have on file. Not a number provided in the email or text. Not a callback to the number that just called you. A number from your contacts, your vendor file, or your bank's official website.

Write this down as a policy. Post it near the computer of anyone who handles payments. Make it non-negotiable. Every legitimate business partner will understand a 60-second callback for security purposes. Anyone who pushes back or creates urgency around skipping verification is telling you something important.

2. Set up a code word system for high-stakes verification

For situations where you need to verify identity over the phone or video, establish a rotating code word or phrase known only to your core team. If someone calls claiming to be you or another team member and asks for something unusual, the employee asks for the code word. A legitimate caller knows it. An AI-cloned voice does not.

This is especially important for childcare centers. Set a parent verification code word for any pickup or payment change request. It's a simple step that eliminates most social engineering attempts targeting your staff.

3. Enable multi-factor authentication on everything, not just email

MFA is table stakes, but many small businesses only enable it on email and leave everything else unprotected. Your accounting software, payroll system, banking portal, Microsoft 365 admin account, and any system with financial data needs MFA enabled. Use an authenticator app rather than SMS codes — SIM-swapping attacks can intercept text messages.

If your banking portal still allows transfers after only a username and password login, call your bank this week and ask about their enhanced security options. Most banks offer callback confirmation or dedicated secure messaging for wire transfers that costs nothing to enable.

4. Use AI-powered email security, not just spam filters

Traditional spam filters work on known signatures and blacklists. They cannot catch novel AI-generated phishing messages that contain no malicious links and no blacklisted domains. Modern email security tools from vendors like Microsoft Defender for Business, Proofpoint, or Mimecast use behavioral analysis to flag unusual sender patterns, look-alike domains, and anomalous email flow even when individual messages appear clean.

If your current email security is just whatever came bundled with your hosting plan, it's almost certainly inadequate for 2026 threats. This is one area where upgrading your toolset makes a measurable difference.

5. Train your team with realistic AI phishing simulations

Annual security awareness training that shows employees what 2019 phishing looked like does not prepare them for 2026 attacks. Current training should include simulated AI-generated phishing emails that target each employee personally, using their real job title, actual colleagues' names, and current projects. The goal is not to trick employees, but to build reflexes so that personalized urgency triggers skepticism rather than compliance.

Platforms like KnowBe4 and Proofpoint Security Awareness offer simulated phishing at a scale that works for small teams. The best programs send two to four simulated attacks per employee per month and track click-through rates over time so you can see whether training is actually working.

6. Review your cyber insurance for social engineering fraud coverage

Many standard cyber insurance policies cover ransomware and data breaches but classify AI phishing, BEC, and deepfake fraud as "social engineering" losses — which may have separate, lower sublimits or specific conditions that must be met before a claim is paid. Some policies require that you attempted to verify the transaction through a specific documented process.

Pull out your current policy and look specifically for the social engineering fraud section. If your coverage is less than $250,000 for this category, discuss adding a crime or fraud endorsement with your broker. Premiums for these endorsements are generally reasonable for small businesses, and a single successful BEC attack routinely costs far more.

What a security incident response looks like for a small business

If someone on your team realizes they may have fallen for one of these attacks, speed matters enormously. The goal is to interrupt the money movement before it clears or to lock down the compromised account before the attacker does more damage.

The immediate steps: Call your bank's fraud line directly (use the number on the back of your card or the official website) and report a potentially fraudulent wire transfer. Banks can sometimes recall wires within the first few hours if they haven't yet settled. Simultaneously, change the password and revoke sessions on any compromised account. If an email account was breached, check for forwarding rules the attacker may have set up to keep reading your mail after you change the password.

Then call your managed IT provider or cybersecurity contact. They can audit your email environment for indicators of compromise, check for lateral movement if credentials were stolen, and help you file the required reports. In California, a breach affecting more than 500 residents triggers notification obligations under state law. Your IT partner should know the requirements and timelines.

What EDCON does to protect small businesses from AI fraud

EDCON's cybersecurity work with small businesses and childcare centers in Southern California starts with an honest assessment of current exposure. Most businesses we talk with have the basics in place — Microsoft 365 with MFA, some form of backup — but significant gaps in email security, employee training cadence, and incident response planning.

On the technical side, EDCON deploys and configures Microsoft Defender for Business, which includes AI-powered email threat protection, endpoint detection, and identity threat monitoring under a single license that makes financial sense for teams of five to 50. For businesses that already have this in place, we often find the most valuable configuration options are not enabled by default and require deliberate setup — sender authentication enforcement, advanced anti-phishing policies, and real-time alerting to a monitored inbox.

For childcare centers specifically, EDCON addresses the physical dimension of these threats: configuring SenLobby.ai visitor management so that pickup authorization is digitally verified against pre-registered contacts, rather than relying on a staff member's recognition of a voice or face. When identity verification is built into the process rather than left to individual judgment, cloned voices and deepfake calls become much less effective.

On the human side, EDCON sets up phishing simulation programs for client teams and runs quarterly security check-ins to review new threats and reinforce correct procedures. The goal is not to make employees paranoid, but to give them a clear, simple process they trust and follow consistently: verify before you act. That one habit, applied reliably, stops most AI fraud attempts regardless of how sophisticated they become.

We also help clients document their verification procedures, incident response steps, and insurance requirements so that if something does go wrong, they have the records they need to file a claim and recover as quickly as possible.

Frequently asked questions