By using this site, you agree to the Privacy Policy and Terms of Use.
Accept
Logic & LayersLogic & Layers
  • Tools
  • Earn with AI
  • Productivity
  • Automation
  • Guides
Logic & LayersLogic & Layers
  • Privacy Policy
  • About
Search
  • Tools
  • Earn with AI
  • Productivity
  • Automation
  • Guides
  • About
  • Contact
  • Blog
  • Privacy Policy
  • Complaint
  • Advertise
© 2026 Logic and Layers. Ruby Design Company. All Rights Reserved.
Person reviewing a suspicious invoice email on a laptop screen
Automation

AI-generated BEC scams: how to spot a fake invoice email

Editorial Team
Last updated: September 14, 2026 2:30 pm
Editorial Team
Share
Microsoft researchers documented a million-email BEC campaign built with AI help.

An email lands in your accounts payable inbox. It’s from your CEO, it references a real software vendor, it includes an invoice for $49,000, and it even shows a forwarded conversation where your boss and the vendor’s leadership discuss the rollout. It’s completely fake. AI-generated BEC scams like this one just hit more than a million inboxes in a single campaign, and most of them looked this clean.

Contents
What happened in the campaign Microsoft just exposedWhy AI-made scam emails are harder to spotInside the fake emailThe 7 signs that give these emails awayWhat to do before you pay any invoiceWhere to report it if money movesFree ways to protect your business

What happened in the campaign Microsoft just exposed

Microsoft Security Research documented one of the clearest AI-generated BEC scams on record: a business email compromise campaign that delivered over a million phishing messages between August 3 and August 5. Nearly 88% of the targeted organizations were in the United States, mostly in IT services, business advisory, and consumer goods.

The play was straightforward and patient. Attackers bought lookalike domains days in advance, impersonated executives, and posed as a well-known vendor. The fake ServiceNow subscription invoice asked for an ACH payment of almost $50,000 per attempt, and it went straight at the people who actually cut checks.

Microsoft’s analysts noted that this wasn’t the usual one-lure scam. The campaign layered executive impersonation, vendor branding, fabricated invoices, and supporting email conversations into one story designed to lower your guard. Each email was personalized too, with the recipient company’s name and an executive’s name filled into the “BILLED TO” section.

Why AI-made scam emails are harder to spot

Old phishing was easy to catch. Broken grammar, weird greetings, a sense of urgency you could smell through the screen. That era is ending.

Researchers found clear fingerprints of generative AI in the email templates: extensive HTML comments, neatly labeled sections, and highly uniform layout syntax across messages sent to completely different companies. The machine wrote clean code, in other words. Ironically, the AI even inserted em dashes and repeated decorative banner lines, the same tics writing detectors flag in blog posts.

The result is an invoice email that reads, formats, and signs like the real thing. Standard advice like “watch for typos” is now nearly useless. AI-generated BEC scams shift the challenge from proofreading to verification, which is exactly where the next section comes in.

Inside the fake email

It helps to see how the pieces stack up, because the scam only works when everything looks connected.

The sender display name shows your CEO. A reply-to address points at a lookalike domain, and the signature repeats the executive’s name and title. Next, the body presents an annual subscription invoice from ServiceNow, a vendor your company plausibly uses. Attached is a detailed invoice for roughly $50,000 with your company’s name filled into the “BILLED TO” section.

Then comes the clever part: a “forwarded” thread where your CEO and ServiceNow’s leadership discuss implementation details. It reads like context, but it’s stage design. The entire goal is to make the accounts payable team feel like the last step in a process that everyone above them already approved.

Now you know why even careful people fall for this. It’s not one lie. It’s a whole paper trail.

The 7 signs that give these emails away

Microsoft’s team flagged the anomalies that exposed these AI-generated BEC scams. Train your team on these seven.

  1. Missing email headers in the “forwarded” thread. Real forwarded emails carry MIME transit headers. These fakes had a clean, headerless block that just looked like a forward.
  2. Flat reply formatting. In genuine threads, nested responses indent. This campaign left them aligned with everything else, because the scammers built the visual thread by hand.
  3. Logic contradictions. The signature instructed recipients not to CC the sender, yet the included thread showed the sender being CC’d the whole time. Machines copy the style of a conversation, not always its logic.
  4. New-domain vendor addresses. The lookalike domains here were registered days before the blast, like service-nowinc.com. Check domain age before trusting any vendor invoice email.
  5. Executive name in the display name, reply-to, and signature. Impersonating a CEO in three places at once is a scam signature, not a flex. Real internal emails rarely need that much authority stacked up.
  6. A payment request routed to new bank details. Multiple banks appeared across the samples, meaning payment destinations shifted between targets. Any invoice with changed banking info deserves a phone call.
  7. Urgency wrapped in normalcy. The fake thread discussed “invoice processing and implementation details” like a routine project. Pressure plus routine is the classic BEC recipe, AI or not.

What to do before you pay any invoice

A seven-point checklist only works if your process enforces it. So make these three rules non-negotiable in your business.

First, verify payment changes out of band. Call the vendor using a number from your records, not one printed in the email. Every single time bank details change, no matter who appears to be asking.

Second, separate approval from payment. The person who approves an invoice shouldn’t be the person who releases the funds. This one control kills most executive impersonation attempts, because a lone “CEO approval” email stops being sufficient.

Third, make the phone call the company policy. If an email requests an ACH payment above a set threshold, a two-minute voice confirmation with a known contact becomes mandatory. For example, AI-generated BEC scams collapse at step one when your process forces the call, since a real ServiceNow rep would have flagged the fake subscription instantly. The FBI logged more than $2.9 billion in BEC losses in a single year, and almost every case traces back to a skipped verification call.

Where to report it if money moves

Speed matters more than embarrassment if a fake payment slips through. Contact your bank the moment someone suspects fraud, because early reports give the bank its best shot at freezing or recalling the transfer. Then file a complaint with the FBI’s Internet Crime Complaint Center at ic3.gov, which is the official channel for BEC cases in the US.

Inside the company, preserve everything: the original email, the headers, and the invoice. Your IT team or security provider can use those details to block the sender infrastructure for the rest of your industry. And drop the shame. These campaigns are built by professionals to fool professionals, and reporting fast is the move that saves the money.

Free ways to protect your business

You don’t need an enterprise security budget to blunt these attacks. Microsoft’s own guidance, published alongside the campaign analysis, boils down to layered basics: phishing-resistant multi-factor authentication, verified contacts for payment approvals, and training that uses real examples like this campaign instead of stale ones. Data hygiene counts too, so it’s worth a reminder about which files should never get uploaded to ChatGPT in the first place.

Set a monthly fifteen-minute drill. Pull the latest scam story, show your team the actual tells, and quiz them. We cover this ground in our AI agent security guide, and the same habit loop works for consumer scams like the fake Amazon messages our Alexa scam checker post explains. Your phone plan, your invoicing software, and your finance team already cost you enough. The rest of this defense is habit, and it’s free.

The bottom line: AI-generated BEC scams prove that AI now writes scam emails better than most humans do, but it still can’t fake a phone call. Verify payments by voice, separate approvals, and teach your team the seven signs above before the next million-email campaign lands in your industry.

You Might Also Like

How to build safe and trustworthy AI agents with Zapier (Beginner guide)
ChatGPT Codex Voice + Sites + Heartbeats: The Voice-First Automation Stack
PCB Design with Claude: AI-Assisted Hardware Design for Beginners
Mistral OCR 4.1: extract text from any document with AI
Connect Google Sheets to ChatGPT: 3 Ways (Pick the Right One)
TAGGED:ai securityBEC ScamsEmail SecurityInvoice Fraudphishing
Share
Previous Article Fathom meeting notetaker interface showing a recorded call with AI notes Superhuman buys Fathom: your meeting notes get an AI agent
Next Article Siri search bar on macOS showing the Ask menu with ChatGPT as the selected model Use ChatGPT or Claude with Siri in iOS 27: full setup guide
Leave a Comment

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

banner banner
Create an Amazing Newspaper
Discover thousands of options, easy to customize layouts, one-click to import demo and much more.
Learn More

Latest News

Tobi Lutke, Shopify's chief executive, speaking at a public event
AI slop is making your job harder, not easier
Productivity
mini-AGI repository card showing a neural network that assembles itself during training
Train your own AI model on one GPU with mini-AGI
Tools
Open-weight AI image generation workflow with transparent layers on a desktop editor
Qwen Image 2.1: free AI images with real transparency
Tools
Illustration of a browser cookie linked to shopping, medical, and debt websites through an ad tracker
ChatGPT ad tracking: how to turn it off in 2 minutes
Tools

Recent Posts

  • AI slop is making your job harder, not easier
  • Train your own AI model on one GPU with mini-AGI
  • Qwen Image 2.1: free AI images with real transparency
  • ChatGPT ad tracking: how to turn it off in 2 minutes
  • Apple AI server: what the M8 Ultra report means for you

Recent Comments

  1. AI slop is making your job harder, not easier on Shadow AI: what it is and how to use AI at work safely
  2. Train your own AI model on one GPU with mini-AGI on Local LLM Hardware Guide 2026: DDR5 Prices Up 500% – What to Buy Instead
  3. Qwen Image 2.1: free AI images with real transparency on GGUF vs GPTQ vs AWQ vs EXL2: pick the right model format
  4. ChatGPT ad tracking: how to turn it off in 2 minutes on Files you should never upload to ChatGPT
  5. Apple AI server: what the M8 Ultra report means for you on Local LLM Hardware Guide 2026: DDR5 Prices Up 500% – What to Buy Instead

You Might also Like

Two workers reviewing the cost of automating their desk tasks with AI
Automation

When AI automation is actually worth it (MIT’s research)

Editorial Team
Editorial Team
9 Min Read
Zapier AI model flexibility - switch between AI models in workflows
Automation

Zapier AI Model Flexibility: Switch Models Without Rebuilding Workflows

Editorial Team
Editorial Team
8 Min Read
MCP servers diagram showing AI connecting to filesystem, Git, web fetch, memory, and Zapier
Automation

MCP Servers You Can Run Today: 5 Copy-Paste Examples for Beginners

Editorial Team
Editorial Team
9 Min Read
//

We influence 20 million users and is the number one business and technology news network on the planet

Quick Link

  • PRIVACY NOTICE
  • YOUR PRIVACY RIGHTS
  • INTEREST-BASE ADSNew
  • TERMS OF USE
  • OUR SITE MAP

Support

  • ADVERTISE
  • ONLINE BESTHot
  • CUSTOMER
  • SERVICES
  • SUBSCRIBE

Categories

  • Tools
© 2026 Logic and Layers. All Rights Reserved.