It is 10:40 on a Tuesday morning, and a small business in downtown Buffalo has a problem: the appointment booking page shows an error, and email is bouncing. The owner is at the counter while customers call to confirm appointments they can no longer make. Nobody is panicking yet, but everyone can feel the morning slipping away.
That is what an IT outage looks like from the inside: customers notice before you do. And here is the uncomfortable part: the symptoms you are seeing — email down, booking down, payments failing — do not tell you what caused them. The same screen can mean three very different things:
- A cyberattack. Someone is targeting your email, or the provider that runs it is under attack.
- A provider outage. The company hosting your email or booking system is down, and you are one of many affected customers.
- A configuration error. A bad update, an expired certificate, or a misconfigured rule broke something that was working fine.
You cannot tell which of the three it is from the error message. That is why the rest of this post is about the habits that make all three outcomes less painful.
If you run a small business in Buffalo or Western New York, 2026 has probably felt like a lot of that. You depend on systems you do not control: email, scheduling, payments, your website, the software your vendors run. When one fails, you find out the way your customers do. This year, AI made some cyberattacks faster and more convincing, and a surprising share of the downtime we saw was not an attack at all, but a change that went wrong.

What AI changes for the people launching attacks
In plain English, AI does not give attackers magic powers. It removes the effort that used to slow them down. Three changes matter most for a small business:
- More convincing impersonation. A phishing email used to give itself away: bad grammar, a generic greeting, a story that did not fit. AI-written messages read like the vendor you actually use, or the “CEO” you actually report to.
- Faster output. What used to take a scammer an hour to write for one target, AI produces in minutes — hundreds of variations, each tuned to a different role.
- Tailored to your business. An attacker who has read your website or your social profiles can address the right person with the right concern: a supplier invoice, a booking-system password reset, a payment-method change.
None of this requires the attacker to be a genius. It requires them to be patient, and AI removed most of the patience.
What your team can recognize and verify
You cannot out-write AI. But you can out-verify it. Three habits catch most of these messages:
- The message creates urgency and asks for action: wire the money, reset your password, approve the change, click the link — today.
- It arrives through a channel you did not expect, or asks you to log in somewhere new.
- The contact details inside the message are not ones you already have.
The rule that stops the most damage costs the least: verify out of band. Call the person or company on a number you already have on file — never the number, link, or reply address inside the message that is asking you to act. CISA’s guidance on multi-factor authentication makes the same point: even a strong password can be phished, which is why the second factor has to be one a fake website cannot copy.
Where AI helps the people defending
The same technology works in reverse, and it is already inside the tools small businesses use. Where defensive AI is genuinely useful today:
- Grouping related alerts. Instead of 300 separate notifications about failed logins, you get one summary: 47 failed logins from one address over 20 minutes, then one success. That is an incident, not noise.
- Flagging unusual sign-ins. A login from a city you have never been to, a device you do not recognize, activity at 3 a.m. on an account that usually moves by day.
- Giving IT staff a starting point. A ranked list of what to look at first, instead of a wall of logs. For a business without a full-time security person, that triage is the difference between acting in an hour and acting in a week.
But there is a limit, and it matters. Defensive AI makes recommendations, and some recommendations are destructive. A tool that is “confident” an account is compromised and locks it out during your busiest afternoon has created an IT outage while trying to prevent a cyberattack. Any action that could lock out an employee or interrupt business — an account lock, a blocked rule, a forced reset — should be confirmed by a person who knows what the business is doing that day. The goal is not to remove the human, but to make the review faster and better informed.
The third cause nobody talks about: good people, bad changes
Here is what the headlines skip. A large share of small-business downtime in 2026 was not an attack at all. It was a change. Three flavors show up again and again:
- A faulty update. A vendor pushes an update to a booking plugin, a payment gateway, or an email rule, and it breaks on your setup. The vendor’s system is fine; yours is not, because yours is the one place that combines their software with yours.
- An expired certificate. The security certificate on your website or mail server quietly expires, and browsers and mail clients start refusing to connect. No attack, no error in the vendor’s dashboard — just a date that passed.
- A misconfigured access rule. Someone tightens a firewall or an email filter to stop spam or a scam, and the new rule blocks your own staff, your customers, or your payment processor.
And here is the part that matters: the person who made the change usually did the right thing. They were following the vendor’s instructions, or fixing a real problem. The gap was not competence. It was process. Most small businesses do not have a staging environment, so the change goes live on the one production system, often at the worst time, with no tested way to undo it. The mistake spreads because nothing was in place to stop it.
That is a fixable gap, and it does not require a security team. Three practices keep a bad change small:
- Staged rollouts. Change one machine or one service first. Watch it for a day. Then the rest. A bad update that hits one register is an annoyance; the same update on every system at once is an outage.
- Peer review. A second set of eyes before the change goes live — even a five-minute “what could this break?” conversation catches a surprising number of mistakes.
- A tested rollback. You know how to undo the change because you have actually undone it in a test. A rollback you have never run is a guess, and the first time you run it is the worst time to learn it.
This is the same logic CISA and NIST apply at scale — NIST’s Cybersecurity Framework is built around understanding and managing cyber risk rather than hoping for perfect prevention — and it scales down to a five-person office.
A one-week checklist: who does what, and how you verify it
Four items. Each has an owner and a way to prove it worked.
- Phishing-resistant MFA on email and admin accounts. Owner: whoever controls the email admin console — or us, if you do not have one. Turn on FIDO2/passkey or hardware-key MFA on the owner’s email, the admin account, and any account that can reset other passwords, wherever the provider supports it. CISA calls phishing-resistant MFA the standard to aim for, because it blocks the fake-login-page trick that defeats text-message codes. Verify it two ways: sign in from a new device and confirm the second step works, then have a colleague try to reset a password and confirm they cannot.
- Verify payment changes by phone. Owner: whoever processes payments. Post one rule where the team will see it: any change to a bank account, a vendor invoice, or a payment method gets confirmed by calling a number you already have on file — never by replying to the message that announced the change. Verify it with a dry run: one person sends a fake “invoice change” email to a colleague, and the colleague calls before acting.
- Test-restore a backup. Owner: your IT person, or us. Pick one set of critical files and actually restore them to a test machine. CISA’s ransomware guidance is blunt about this: keep offline, encrypted backups, and test them regularly. A backup you have never restored is a hope, not a plan. Verify by opening the restored files — and time the whole process. That number is your real recovery time, and it is the number you were guessing at before.
- An offline contact list for critical vendors. Owner: the owner. A printed list of support numbers for your email provider, your payment processor, your booking system, your internet provider, and your IT person — stored somewhere that does not depend on the internet. Verify it the only way that counts: when something breaks, can you find your email provider’s support number without opening email? If not, the list is not done.
What would make 2027 better
Nobody should promise you a calm 2027, and we will not. The attackers will keep getting faster, and some of the outages you experience next year will still be someone else’s bad update. But three things can change the shape of the year, and all three are within your control:
- Fewer routine changes that turn into emergencies. When updates are staged, reviewed, and rolled back, the bad ones stay small: a fifteen-minute blip on one machine instead of an all-day outage.
- Faster recovery when something does fail. A tested backup restore and an offline vendor list turn a six-hour mystery into a one-hour fix, because the first hour is where most of the damage happens.
- Defensive AI doing the tedious work. Alert triage, anomaly flagging, first-pass investigation — the work that used to drown a small IT person — becomes something the software does and a human confirms.
We would frame that as an outlook, not a guarantee. No setup prevents every attack, and no process prevents every outage. What you can control is how small the bad day is, and how fast you are back on it. That is a better question to bring into 2027 than whether you will have a bad day.
Frequently Asked Questions
Short answers to the questions we most often hear from small-business owners after a bad day.
-
Are AI cyberattacks a real threat to a small Buffalo business, or mostly hype?
Real, and they arrive through the same door as always: a person clicking or calling. AI makes the messages more convincing and cheaper to produce, so expect more of them, not fewer. The risk is not a dramatic hack — it is a convincing email that gets one busy person to act.
-
Will phishing-resistant MFA stop every account takeover?
No — nothing stops everything, and we will not promise that. But it closes the most common path: the fake login page. CISA calls FIDO-based, phishing-resistant MFA the standard to aim for, because a passkey or security key only works for the real website it was enrolled with. Any MFA is better than none.
-
Our email went down for an hour. How do we know if it was a cyberattack?
Usually, you cannot tell from the symptoms alone — and that is the point of this post. Provider outages, bad updates, expired certificates, and attacks can all look like “email is down.” Check the provider’s status page, look at your own change history for the day, and write down what happened. The recovery plan should not depend on guessing the cause.
-
What is the single highest-value thing we can do this month?
Test-restore a backup. Pick one set of critical files, restore them to a test machine, open them, and time the process. That one drill tells you your real recovery time, exposes the gaps in the backup, and costs an afternoon. Everything else on the checklist builds on knowing your numbers.
