Table of Contents
Most breaches don’t start with a hooded genius in a basement. They start with a reused password, a contractor’s unpatched laptop, or an invoice that looks exactly like the last forty invoices the finance team approved. The gap between how attacks are imagined and how they land is where small and mid-sized businesses keep losing money, customers and sleep.
What follows are the seven threats that turn up again and again in incident reports, cyber insurance claims and post-mortems. Not an exhaustive list, but the ones worth understanding properly instead of skimming past on a compliance slide.
1. Double-Extortion Ransomware
Ransomware stopped being about locked files around 2020. Modern crews copy the data first, encrypt second, then threaten to publish what they took unless they get paid twice: once for the decryption key, once for silence. Ransomware-as-a-service made this cheap to enter. A would-be criminal can rent a ready-built payload, a leak site and a negotiation playbook for a cut of the proceeds, which means the people attacking your file server may have no technical skill at all.
Two things catch companies out. The first is backups that exist but have never been restored. Ask a blunt question in your next management meeting: when did we last rebuild a full system from backup, and how many hours did it take? If nobody knows, that’s your answer. The second is dwell time. Attackers routinely sit in a network for weeks mapping domain controllers and finance systems before detonating anything, so the encryption you see on a Monday morning was probably planned the previous month.
Paying rarely ends the relationship
A payment buys a promise from people whose business model is breaking promises. Several well-documented cases involve organisations paying once, then being hit again months later because the stolen data never went away. Insurance carriers now ask harder questions about segmentation and offline backups before they’ll even quote a policy.
2. AI-Assisted Phishing and Business Email Compromise
Spear-phishing used to betray itself with clumsy grammar. Language models removed that tell. A convincing email now arrives in fluent, correctly-punctuated prose that mimics the chief financial officer’s tone, references a real project and asks for a supplier’s bank details to be updated before close of business.
The deepfake call is no longer science fiction
In 2024, an employee at the engineering firm Arup transferred roughly $25 million after a video conference in which every other participant was a deepfake. Voice cloning is even easier: fifteen to thirty seconds of a manager’s voice from a webinar recording is enough. The practical defence is boring and effective. Any change to payment details gets verified by calling a number you already hold, not one supplied in the email. Two people, two channels, no exceptions.
AI capability is improving faster than most security teams can adapt, which is one reason so much capital keeps flowing toward the AI stocks investors are chasing. The same capability lands in an attacker’s hands within months, usually for free.
3. Supply Chain and Third-Party Compromise
You inherit your vendors’ security posture whether you like it or not. SolarWinds in 2020, the 3CX installer in 2023 and the MOVEit file-transfer flaw later that same year all followed the same pattern: one trusted supplier, thousands of downstream victims, no phishing required.
The realistic response for a mid-sized company isn’t to audit every vendor. It’s to know which three or four suppliers could take you down if they were compromised, ask them for a recent independent security report, and check the breach notification clause in their contract. If a vendor holds customer data and the contract gives them sixty days to tell you about a breach, that clause is worth renegotiating.
4. Unpatched Software, End-of-Life Systems and Shadow IT
Every organisation has a Windows Server 2012 box that nobody remembers installing, quietly holding the payroll database. Windows Server 2012 lost official support in October 2023, which means every flaw discovered since then stays open. The same is true of forgotten VPN appliances, out-of-support firewalls and the ageing NAS drive in a branch office.
Shadow IT makes the inventory problem worse. A marketing team signs up for a design tool, connects it to the customer list through a personal account, and nobody in IT knows it exists. Keeping an eye on which tools are actually spreading through teams is part of staying ahead; trend breakdowns such as O’Reilly’s technology trends for 2025 are a useful early-warning system for what your staff will adopt before security reviews it.
5. Stolen Credentials and Account Takeover
Stolen logins remain the single cheapest route into a business. Infostealer malware scrapes saved passwords and active session cookies from browsers and sells the bundles in bulk. That detail matters: stealing a live session cookie lets an attacker skip multi-factor authentication entirely, because the system thinks the session was already verified.
Browser-based tools make this surface bigger than most people assume. Business Chrome extensions can save a team hours a week, but any add-on with permission to read and change data on every page is effectively running with your users’ trust. Audit them the way you’d audit installed software, and remove what nobody recognises.
- MFA fatigue attacks work by spamming push notifications until someone taps approve out of frustration. Uber’s 2022 breach started exactly this way.
- Passkey and hardware-key authentication defeats both phishing proxies and push bombing, because the credential is bound to the real domain.
- Session revocation after a suspicious login is a checkbox most admin consoles hide in a submenu. Find it before you need it.
6. Insider Risk, Both Careless and Deliberate
Not every incident involves an outsider. The far more common version is a departing employee who downloads a shared folder on their last Friday, or a finance team that has edit rights on the whole drive because restricting it felt awkward in 2019. Over-permissioned accounts turn a single compromised login into a full data loss event.
Deliberate insider theft tends to cluster around resignations, pay disputes and redundancies. You don’t need a forensic team to handle it. You need an offboarding checklist that runs the same day, least-privilege access by default, and logging that shows who opened what and when.
7. Cloud Misconfiguration and Exposed Data
Cloud providers secure the building; you secure what you put inside it. A storage bucket switched to public takes one setting, and it can expose years of customer records. The same pattern repeats across over-permissive identity roles, forgotten development environments holding a copy of production data, and APIs with no rate limiting at all.
Configuration drift is the real enemy. An environment that passed a review in January has had two hundred changes applied since. Automated scanning, infrastructure-as-code checks and a rule that every new cloud account starts from a hardened template close most of the gap. Cost is not the blocker here; process is.
What to Fix First When the Budget Is Small
Security vendors would like you to believe you need a platform. In practice, a mid-sized business that nails the fundamentals is a harder target than most attackers want to bother with, and they’ll move to a softer one.
- Get an accurate asset list, including the machines nobody owns and the SaaS tools nobody approved. You cannot patch what you can’t see.
- Patch the internet-facing stuff first, within days rather than quarters, and isolate anything that’s out of support.
- Move to phishing-resistant authentication on email, finance systems and remote access. This is the highest-return change on the list.
- Test a backup restore this quarter and keep one copy offline or immutable.
- Write down the payment-change rule and get it signed by whoever approves invoices.
- Run a one-hour tabletop exercise. Ask what happens in the first sixty minutes after someone notices odd file extensions on the shared drive. The gaps that surface will be more useful than any report.
Watching which technologies are about to land in your environment helps too. Forecasts like O’Reilly’s Signals for 2026 are worth reading not because every trend matters, but because the ones that do will show up in your staff’s workflows long before they show up in a policy document.
Pick two items from that list and finish them. A patched estate, tested backups and a phone call before any bank details change will stop more real-world attacks than a dashboard nobody reads.


