An email attack only works if a human is in the loop

A phishing attack only works if a human clicks the link. A malicious HTML payload only works if a mail client renders it. A tracking pixel only works if an image loader fetches it. Display-name spoofing only works if a human reads “CEO” and skips the actual sender address. Homograph domains only work if a human glances at paypaI.com and sees paypal.com. An attached macro only works if someone double-clicks the file.

Look at the pattern. Every common email attack requires the human to handle raw email. The Verizon DBIR put a human in the loop on 68% of all breaches in 2024, and email is where most of those humans got reached.

The attack surface is the human-in-the-loop. Email is just the delivery van.

Twenty years of security tooling has been trying to protect that human from a hostile data stream they were never qualified to parse. Spam filters. Link rewriters. Image proxies. URL detonation sandboxes. “Did you really mean to send this to an external recipient?” banners. It’s all the same shape, guardrails bolted onto an architecture that put a non-technical user in front of an internet-facing inbox and asked them to make judgment calls about parsed MIME at 7am.

Take the human out of that loop and most of those attacks die. The attacker did not get worse. The surface they were attacking no longer exists. An AI reads the email, names the actual sender, calls out the urgency manipulation, quarantines the attachment, and hands the user a sentence in plain English. The hostile data stream never reaches a person.

Security is mostly a UX failure. Most of what we call “attack surface” is just the set of places we made humans look at raw data we already knew they couldn’t safely interpret.

More training does not fix this. The only fix is to stop putting them there.