The ten biggest cyber risks for small businesses in the age of AI

Threats · November 2026 · About a 10 minute read

AI hasn't invented many new attacks. What it has done is make the old ones cheaper to run, faster to launch and much harder to spot. Small and medium businesses feel that most, because they usually have the fewest people and tools to defend themselves.

We went through this year's main reports, including Verizon's Data Breach Investigations Report, the FBI's internet crime report and the UK government's Cyber Security Breaches Survey, alongside research on phishing, deepfakes and vulnerability exploitation. Here's what they show, in order of how often we think they'll matter to a typical small business, and what to do about each one.

1. Phishing that no longer looks like phishing

The old advice was to look for bad spelling. That doesn't work anymore. Microsoft's research found that AI-written phishing emails got clicked far more often than traditional ones, at around 54% against 12% in its tests. In the UK, phishing is still the most common attack businesses report.

What to do: set up SPF, DKIM and DMARC so nobody can send email as you, use good email filtering, and train staff on what modern phishing looks like. And make reporting a suspicious email easy and welcome.

2. Fake invoices and changed bank details

Business email compromise, where criminals pose as a supplier, a customer or your own boss to redirect a payment, cost victims more than $3 billion in 2025 according to the FBI. It's one of the most expensive crimes there is, and it doesn't need any malware at all.

What to do: never act on a change of bank details or an urgent payment request without confirming it through a separate channel, using contact details you already hold. Put that rule in writing.

3. Deepfake voices and video calls

Criminals can now clone a voice from a short recording and use it to call your finance team "from the CEO". This isn't only aimed at big companies. Research from the identity firm Sumsub found that a large share of deepfake attacks target businesses with fewer than 50 employees, and money sent this way is rarely recovered.

What to do: the fix is process, not technology. Agree a call-back rule and a code word for payment requests, and require two people to approve larger payments.

4. Known software flaws, exploited fast

For the first time in the history of Verizon's report, exploiting a software vulnerability overtook stolen passwords as the most common way attackers got in, at 31% of breaches where the entry point was known. Attackers are also getting quicker. VulnCheck found that almost a quarter of exploited vulnerabilities were attacked on or before the day they were made public.

What to do: know everything you have facing the internet, and patch the flaws that are actually being exploited first. CISA's Known Exploited Vulnerabilities list is a free and reliable guide to which those are. Waiting for an annual check is no longer enough.

5. Phishing kits that get past sign-in codes

Two-step sign-in is still essential, but criminals now sell kits that sit invisibly between your staff and the real Microsoft or Google login page and steal the session after the code has been entered. In 2026 the FBI publicly warned about one such kit aimed at Microsoft 365. Small businesses are exposed because most rely on app codes or text messages, which these kits are designed to defeat.

What to do: use phishing-resistant sign-in, such as passkeys or security keys, at least for admins and finance staff. Turn on conditional access rules, and block sign-in methods you don't use.

6. Password-stealing malware

Infostealers are malware that quietly copies saved passwords and logged-in sessions from a browser. Outpost24's 2026 research found credentials stolen this way linked to 72% of the organisations it examined. One popular trick is the fake "verify you're human" page that tells you to paste a command into your computer. ESET reports these attacks have more than doubled in a year.

What to do: use a proper password manager rather than saving passwords in the browser, keep endpoint protection running, and teach one simple rule: a real CAPTCHA never asks you to paste or type a command. If a device is infected, sign the user out of every session, not just reset the password.

7. Ransomware

Ransomware appeared in a far higher share of breaches at small and medium businesses than at large ones in Verizon's 2026 data. The good news is that backups work: Verizon reports that many smaller businesses refused to pay because they could restore their data themselves.

What to do: keep backups that can't be changed or deleted from your normal network, and test restoring from them. Patch quickly (risk 4), protect sign-ins (risk 5), and have a plan (risk 10).

8. Attacks that come through your suppliers

Verizon found that 48% of breaches involved a third party, up from 30% the year before. Your IT provider, your software vendors and the apps connected to your email all have some access to your business. Yet fewer than a third of UK businesses formally checked their suppliers' security in the past year.

What to do: list the suppliers with access to your systems or data, ask the important ones about their security (a Cyber Essentials or ISO 27001 certificate is a good sign), and give each one only the access it needs.

9. Company data leaking into AI tools

Staff use AI because it saves time. But when someone pastes customer data or a contract into a personal AI account, that information has left your control. Verizon found that the share of employees regularly using AI on work devices tripled in a year, and that unapproved "shadow AI" is now a common source of accidental data leaks.

What to do: don't ban AI. Give people approved business accounts with the right data settings, and a short, clear policy on what can and can't go into them.

10. Your own AI tools being turned against you

If you've added a chatbot to your website or an assistant connected to your files, you've added a new kind of risk. Prompt injection, where hidden instructions in a message or document hijack the AI, has been the top risk on the OWASP Top 10 for LLM Applications in every edition. The more your AI can do (send emails, update records, read your inbox), the more an attacker can do through it.

What to do: give your AI only the access it truly needs, keep a person in the loop for risky actions, and have it tested before and after launch.

And the one that ties them together: no plan

Only about a quarter of UK businesses have a formal incident response plan. Yet the basics make a real difference. UK government data shows that organisations with Cyber Essentials, which covers five basic controls, make 92% fewer cyber insurance claims than those without it.

None of this needs a big budget. Most of it needs a clear picture of where you stand and a sensible order to fix things in. That's exactly what our Rapid Exposure Audit is for.

Want to know which of these apply to you? Book a free 30-minute call, or start with a fixed-price audit.

Sources

  1. Verizon, 2026 Data Breach Investigations Report, news release (May 2026)
  2. UK Department for Science, Innovation and Technology, Cyber Security Breaches Survey 2025/26
  3. FBI Internet Crime Complaint Center, 2025 Internet Crime Report (April 2026)
  4. VulnCheck, State of Exploitation 1H 2026
  5. Outpost24, 2026 Credential Exposure Benchmark, as reported by eSecurity Planet
  6. ESET Threat Report H1 2026 findings on ClickFix, as reported by CybelAngel
  7. FBI public service announcement on the Kali365 phishing kit (May 2026), as reported by Guardz
  8. NCSC, A decade of Cyber Essentials
  9. OWASP GenAI Security Project, Top 10 for LLM Applications
  10. Cloud Security Alliance, research note on AI-weaponized phishing, citing the Microsoft Digital Defense Report 2025
  11. Summary of Sumsub's 2025 deepfake fraud findings

Related reading