How to Protect Your Small Business from AI-Powered Phishing

AI did not invent phishing. It industrialised it. In 2026, attackers can generate fluent, personalised emails that copy your vendor’s tone, reference a real invoice number, and land in inboxes that already passed basic authentication—because the message often comes from a compromised legitimate account. Small businesses are attractive targets: fewer security staff, shared passwords, and owners who approve payments from their phones between meetings.

This tutorial gives you a practical defence system you can implement without a security department: awareness, habits, tools, response playbooks, and safe ways to use AI for triage.

Key takeaways

• AI spear phishing looks personal, urgent, and grammatically clean—grammar mistakes are no longer a reliable red flag.
• Most successful attacks exploit people and payment processes, not exotic malware.
• Train every person who can move money, change payroll, or reset passwords—especially founders.
• Turn on the strong built-in controls in Google Workspace / Microsoft 365 before buying anything fancy.
• Write a one-page incident playbook for a 5-person team and a clearer RACI version for ~25 people.
• You can use AI to help triage suspicious mail, but never paste credentials, full account dumps, or live session tokens into a chat.
• Industry investment is rising fast—for context, email-security startup AegisAI raised a $36M Series A in July 2026 to fight AI spear phishing—which signals how serious the threat has become, not which vendor you must buy.

Who this is for

Owners, office managers, operations leads, and IT generalists at small businesses (about 2–50 people) who use Google Workspace or Microsoft 365. If you process invoices, payroll, wire transfers, or customer PII, this guide is for you.

Prerequisites

1. Admin access (or a partner who has it) to your email workspace.
2. A list of people who can approve payments, change banking details, or reset accounts.
3. A shared password manager (1Password, Bitwarden, etc.) already in use—or a decision to adopt one this week.
4. A communication channel for emergencies that is not only email (phone tree, Signal/WhatsApp group, or Slack huddle rules).
5. 60–90 minutes for the first hardening pass, then 30 minutes for a team training.

Step 1: Understand what AI spear phishing looks like in 2026

Classic phishing was noisy: weird domains, broken English, spray-and-pray. AI-assisted spear phishing is quieter.

Common patterns you will see

1. Vendor payment change. “Updated banking details for invoice #4812—please pay by Friday.” The invoice number may be real.
2. Executive impersonation. A fake founder asks finance to “handle this quietly before the board call.”
3. Payroll redirect. “HR” asks an employee to update direct deposit via a lookalike link.
4. IT / MFA fatigue. Messages that push you to approve a login or “reset SSO now.”
5. Shared-thread replies. Attackers inject into a real conversation using a compromised mailbox, so the thread history looks trustworthy.
6. Callback social engineering. Email starts the story; a phone call “from the bank” finishes the scam.

Why old heuristics fail

• Perfect grammar is normal now.
• Logos and formatting are easy to clone.
• Many attacks pass SPF/DKIM/DMARC because they are sent from real compromised accounts.
• Urgency plus authority still works on busy humans.

Mindset shift: Treat unexpected money, access, or data requests as untrusted until verified on a second channel you already know.

Step 2: Run a 30-minute risk snapshot

Before tools and policies, map your blast radius.

Answer these questions in a shared doc:

1. Who can approve payments over $500 / $5,000 / any wire?
2. Who can change vendor bank details?
3. Who can reset passwords or MFA for other people?
4. Where do invoice PDFs usually arrive (shared inbox? personal founder inbox?)?
5. What is our out-of-band verification channel for payment changes?
6. When did we last review mailbox forwarding rules and OAuth app access?

If you cannot answer #1–#3 in five minutes, fix that first. Attackers love ambiguity.

Step 3: Train every employee with a short, memorable checklist

Long annual security slideshows do not stick. Use a one-page checklist and practice it.

Employee training checklist (print / pin / LMS)

Pause if the message asks you to:

• Pay, refund, or change bank details
• Share passwords, MFA codes, or backup codes
• Click a link to “verify” payroll, SSO, or shipping
• Install software or enable mailbox rules
• Keep a financial request secret from the team

Then do this:

1. Check the true sender address and reply-to—not just the display name.
2. Hover links (on desktop) and compare the real URL to the brand you expect.
3. Open the vendor site from your bookmarks/password manager—not from the email link.
4. Verify payment or data changes on a known phone number or in-person / video call.
5. Report suspicious mail with your workspace’s report button and notify the owner/admin.

Quarterly drill (45 minutes)

1. Send a simulated phishing email (built-in tools or a reputable vendor).
2. Review who clicked, who reported, who ignored.
3. Praise reporters publicly. Coach clickers privately—blame culture hides incidents.
4. Update the checklist with one new real-world example from your industry.

Step 4: Install email verification habits for money movement

Create a two-person, two-channel rule for high-risk actions.

Payment change protocol

1. Email request alone is never enough to change bank details.
2. Caller uses a phone number already on file (not the number in the email).
3. Second approver confirms in writing in your accounting tool or shared finance channel.
4. First payment to new details is capped and watched.
5. Log the verification (who called whom, when).

Everyday verification habits

• Prefer password-manager saved sites over search results and email links.
• Treat “gift cards / crypto / wire today” as almost always hostile.
• Be suspicious of sudden secrecy (“don’t tell accounting”).
• If a VIP asks for something unusual, assume compromise until proven otherwise.

These habits matter more than any single filter.

Step 5: Choose a tool stack (built-in first, dedicated later)

Tier 0 — Free discipline

• Password manager for all staff
• MFA everywhere (prefer phishing-resistant keys / passkeys where possible)
• No shared “team@” passwords in a spreadsheet

Tier 1 — Harden Google Workspace or Microsoft 365

Google Workspace admin priorities:

1. Enforce 2-Step Verification / passkeys for all users.
2. Turn on advanced phishing and malware protections available on your edition.
3. Disable automatic forwarding to external addresses unless explicitly approved.
4. Review OAuth apps and remove stale third-party access.
5. Set alerts for suspicious login events.
6. Use the Report phishing workflow and make sure reports reach an admin.

Microsoft 365 admin priorities:

1. Enforce MFA / Conditional Access baselines.
2. Enable Defender anti-phishing policies appropriate to your license.
3. Turn on mailbox intelligence / impersonation protections if available.
4. Block legacy authentication.
5. Audit inbox rules that forward or delete mail automatically.
6. Make “Report message” easy and monitored.

For many 5–15 person companies, Tier 1 plus habits stops a large share of opportunistic attacks.

Tier 2 — Dedicated email security

Consider a dedicated layer when:

• You process high-value wires weekly
• You are in finance, legal, healthcare-adjacent, or construction/vendor-heavy niches
• You already had a close call
• Your built-in license tier lacks strong anti-phishing features

Options generally include Google/Microsoft higher security SKUs and third-party email security platforms. Evaluate vendors on false-positive rate, admin time, Microsoft/Google integration, and whether they analyse intent—not only known-bad URLs.

Industry context only: In July 2026, AegisAI announced a $36 million Series A (about $49 million total funding) focused on AI-era spear phishing defence. That funding wave is useful context that the category is attracting serious investment. It is not a purchase recommendation. Choose tools based on your stack, budget, and admin capacity.

Tier 3 — Process tooling

• Accounting software with vendor-change approvals
• Shared inboxes for ap@ / billing@ with two readers
• Hardware security keys for admins and finance roles

Step 6: Write incident response playbooks (5-person and 25-person)

Speed beats perfection. Pre-write who does what.

Playbook A — 5-person team (owner-led)

If someone suspects a phishing email:

Minute 0–15

1. Do not click further. Do not enter passwords.
2. Take screenshots of the email (headers if you know how).
3. Report via the email client’s report button.
4. Call/text the owner or admin on the emergency channel.
5. If credentials were entered: change password from a different device if possible, revoke sessions, rotate MFA.

Hour 0–1

1. Admin checks forwarding rules, recent OAuth grants, and sent mail.
2. Search the org for the same subject/sender pattern; remove from inboxes if tools allow.
3. If money may be in motion: call your bank’s fraud line immediately.
4. Notify anyone who might trust the compromised thread.

Hour 1–24

1. Reset affected passwords; force logout.
2. Review payment queue for altered invoices.
3. Write a 10-line incident note: what happened, what we did, what we will change.
4. Do a 15-minute team debrief the next morning.

Roles in a 5-person company: Owner = incident lead. Whoever has admin = technical lead. Bookkeeper = payment freeze. Everyone else = stop clicking and report.

Playbook B — ~25-person team (lightweight RACI)

Define owners in advance:

Severity ladder:

1. S3 — Suspicious email, no click: report, delete/quarantine, optional awareness note.
2. S2 — Click or credential entry: contain account, force resets, search lateral movement.
3. S1 — Money moved or data exfiltrated: bank fraud process, insurance, legal, full timeline.

Keep contact numbers for bank fraud, MSP, cyber insurance, and counsel in the same one-pager.

Step 7: Use AI to triage suspicious mail—safely

AI can help a busy owner decide “ignore, report, or escalate,” but reckless pasting creates new risk.

Safe triage pattern

1. Copy the email body text and subject.
2. Redact customer account numbers, SSNs, passwords, and magic links.
3. Ask an AI assistant:

You are helping a small-business owner triage a suspicious email.
Do not assume it is safe.
Flag: urgency, money requests, credential requests, lookalike brands, odd links, secrecy pressure.
Ask me for any missing header clues.
Recommend: Ignore / Report / Escalate to admin / Call bank.
Email text:
[redacted paste]

4. Treat the answer as a second opinion, not a green light.
5. Still verify high-risk requests out of band.

Never paste into AI chats

• Passwords, MFA codes, session cookies
• Full raw .eml with auth tokens in links if avoidable
• Entire customer databases “for analysis”
• Private keys or VPN configs

If you already use Claude or Gemini for productivity, keep a separate habit: business drafting is fine; security incidents need redaction discipline. For productive AI use after your defences are in place, see How to use Claude Opus 5 for everyday small-business workflows.

Step 8: Hardening extras most SMBs skip

These take an afternoon and close common holes attackers still abuse:

1. Mailbox rule audit. Look for rules that forward externally, delete mail, or hide messages with certain words (“invoice,” “wire,” “payroll”).
2. External banner. Turn on “external sender” warnings so lookalike domains are harder to miss.
3. Admin alerts. Email/SMS when a new forwarding rule appears, MFA is disabled, or a suspicious login succeeds.
4. Vendor callback sheet. Keep a short list of top 20 vendors with the phone numbers you will actually use for verification.
5. Offboarding in 24 hours. Former contractors with lingering mailbox access are a quiet phishing beachhead.
6. Domain lookalikes. Periodically search for domains that mimic yours; at minimum, watch invoice footers and social bios for wrong URLs.

None of these replace training, but they shrink the window between “attack sent” and “someone notices.”

Step 9: Adopt simple policy templates

Copy, adapt, and publish these in your employee handbook.

Policy template 1 — Email and payments

1. No payment destination changes based on email alone.
2. All wires above $[X] require two approvers.
3. Finance verifies vendor detail changes by phone using on-file numbers.
4. Shared finance mailboxes must have at least two active monitors.
5. Violations are process failures to fix—not only individual blame.

Policy template 2 — Accounts and access

1. Unique passwords in the company password manager.
2. MFA required for email, banking, payroll, and admin tools.
3. Admins use separate admin accounts where practical.
4. Offboarding checklist includes mailbox, forwarding, OAuth, and device recovery within 24 hours.
5. Quarterly access review for finance and admin roles.

Policy template 3 — Reporting culture

1. Report suspicious messages within 15 minutes of noticing.
2. Fast reporting is rewarded; honest mistakes are coachable.
3. Do not forward suspected phishing widely (it spreads risk); use official report tools.
4. External “IT support” cold calls are unverified until we call back on a known number.

Real example walkthrough: the almost-wire

A 12-person design studio receives an email that appears to continue a real vendor thread about a $18,400 final invoice. The PDF looks perfect. The reply asks them to use “updated” bank details because of an audit.

What worked:

1. The office manager paused because bank changes are on the training checklist.
2. She did not call the number in the PDF. She called the vendor number in the password-manager / CRM record.
3. The real vendor said no change was requested.
4. Admin searched other inboxes, found two similar messages, and removed them.
5. They filed a short incident note and added “bank detail changes” to the next all-hands (5 minutes).

What almost failed: The founder was travelling and nearly approved from a phone preview without opening headers. The two-person rule saved them.

Total loss: $0. Total time: about 40 minutes. That is a successful security outcome.

Best practices

1. Defend processes first—especially money movement.
2. Prefer passkeys / security keys for admins and finance.
3. Keep emergency comms outside email only.
4. Review forwarding rules and OAuth grants monthly.
5. Simulate phishing quarterly; coach without humiliation.
6. Document vendor verification steps where AP staff can see them.
7. Buy tools only after Tier 1 hardening is done.
8. Redact before AI triage.
9. Put cyber insurance contacts in the playbook before you need them.
10. Rehearse the 15-minute response once so it is not theoretical.

Mistakes to avoid

1. Assuming “we’re too small to target.”
2. Relying on grammar mistakes to spot scams.
3. Letting the founder be the only approver for every wire while also living in email all day.
4. Buying a security product and skipping training.
5. Sharing one mailbox password across interns.
6. Calling phone numbers supplied inside the suspicious message.
7. Quietly paying a “small” test invoice to new bank details “just to unblock.”
8. Pasting full incident artifacts into random AI tools.
9. Punishing reporters who were wrong—next time they will stay silent.
10. Ignoring compromised-lookalike domains on social profiles and invoice footers.

Conclusion

AI-powered phishing succeeds when urgency outruns process. You do not need a giant SOC to fight back. Teach a short checklist, harden the email platform you already pay for, enforce two-channel verification for money and access changes, and keep a one-page playbook for the bad day. Use AI carefully as a triage assistant—not as a place to dump secrets.

Do the risk snapshot and payment protocol this week. That alone prevents the most expensive story your company never wants to tell.

Sources

• AegisAI / PR Newswire, “AegisAI Raises $36 Million Series A Led by Battery Ventures to Fight the New Wave of AI Spear Phishing” (July 23, 2026)
• TechCrunch, “AegisAI, founded by former Google security execs, lands $36M to stop AI-driven spear phishing” (July 23, 2026)
• Company-reported 2025 observations cited in AegisAI materials on AI-generated spear phishing growth and filter evasion (industry context; validate against your own telemetry and primary reports where possible)