Business email compromise is an attacker inserting themselves into a payment conversation you were already having. The invoice looks right because it mostly is.
Short answer: Business email compromise is fraud that works by joining a conversation you were already having about money. Someone gets access to a mailbox, watches until a payment is due, and then intervenes with new bank details. There is nothing malicious to detect in the final message, because the only thing wrong with it is the account number.
Key takeaways
Access. Someone signs into a mailbox using a password from an unrelated breach, or one entered into a convincing login page. Multi-factor authentication, where it is enabled, usually stops this step.
Observation. Nothing happens for days or weeks. The attacker reads. They learn who your suppliers are, what your invoices look like, who approves payments, when the director travels, and how your bookkeeper writes. Often they create a mailbox rule that quietly moves certain replies out of the inbox, so the real supplier's follow up is never seen.
Insertion. When a genuine payment is due, a message arrives about that genuine payment. New bank details, a plausible reason, and correct context. Sometimes it comes from the real compromised mailbox. Sometimes from a lookalike domain a character or two different from the real one.
Payment. The money goes to an account controlled by the attacker and moves onward quickly. Recovery depends on how fast the bank is told, which is why speed matters more than blame in the first hour.
There is no attachment and no link, so scanning finds nothing. The sender is often legitimate, because it really is the supplier's mailbox. The content is not suspicious, because it concerns a real invoice. And the request is not unusual, because suppliers do change bank accounts occasionally.
Every layer of a normal small business email setup is built to detect something bad arriving. Here, nothing bad arrives. The wrongness is in a number that only the supplier can confirm.
Not because they are targeted more thoughtfully. Because of how the work is arranged.
In a firm of eight, the person who receives the invoice is often the person who approves it and the person who pays it. Segregation of duties is a large company idea; here, one person holds the whole chain, so one deception is enough.
Authority is also close and informal. A message that appears to come from the owner carries real weight, and asking the owner to confirm feels like a fuss. Attackers rely on precisely that reluctance.
Finally there is no queue. In a large organisation an unusual payment sits in a process. In a small firm it can be done in ninety seconds, which is not enough time for doubt to arrive.
One rule, written down, no exceptions.
Any change to bank details is verified by voice, on a number you already had, before any payment is made.
The phrase "a number you already had" is the whole control. Not the number in the email, not the number in the new invoice, not the number on the letterhead attached to the message. A number from your own records, or from the supplier's website that you navigated to yourself.
Three things make it hold. Apply it to the owner and the directors, since they are who gets impersonated. Tell your suppliers you do this, so the call is expected and nobody feels accused. And make it apply to first payments to any new supplier, which is when you have no prior details to compare against.
Around it, four supporting measures. Multi-factor authentication on email, which prevents the access step. A periodic check of mailbox rules, because a forwarding or move rule you did not create is one of the few reliable signals of compromise. A second pair of eyes on payments over a threshold you choose. And knowing your bank's fraud line before you need it, because the first hour decides whether funds can be recalled.
Two things have changed and neither is transformational on its own.
Messages read better. Awkward English used to be a tell, and generative writing tools have removed it. The con is unchanged; the polish is better.
Voice is no longer proof. Cloned audio means "I heard them say it" is weaker evidence than it was. This matters for the verification rule, so the call should go to a number you already had, rather than accepting a call that comes to you.
What has not changed is the shape of the fraud or the defence. A verification habit does not care how fluent the message was.
Partly, and it is worth being precise rather than claiming the whole problem.
Prinn does not read your email, does not detect phishing and does not stop a fraudulent invoice. Nothing in it would have prevented the payment in the story above.
What it addresses is one of the supply lines. Business email compromise runs on knowing your suppliers, your clients, your invoice format and your payment timing. When that material is pasted into third party AI tools it accumulates outside your control. Prinn runs on each person's computer and checks a prompt before it is sent; if it finds bank details, card numbers, identity numbers or credentials it shows what it found and offers three choices: redact it, stop, or send it as is. The person decides, and it never blocks anyone. Prompt content never leaves the machine, including to us.
If you are choosing where to spend an hour on this problem, spend it on multi-factor authentication and the payment verification rule first.