How AI-Powered Business Email Compromise Works - and Why DMARC Is Your Best Defense

Business email compromise is not a new threat. What is new is the cost and speed at which attackers can now execute it.

In 2025, researchers at Eye Security studied two commercially available AI-powered phishing kits sold on underground markets. Both kits automated the entire BEC workflow: target selection, email generation, and reply handling. One kit could generate a complete phishing campaign targeting hundreds of recipients in under 30 minutes, using large language models to craft convincing messages tailored to each target's industry and role, then parsing responses to identify promising leads for follow-up attacks.

That is the BEC threat landscape in 2026.

What Is Business Email Compromise (BEC)?

BEC is a fraud scheme where an attacker impersonates a trusted sender - typically a colleague, vendor, or executive - to trick the recipient into transferring money or sharing sensitive information. Unlike phishing emails that blast millions of addresses hoping for a bite, BEC is targeted. The attacker researches a specific organization, identifies a specific person, and crafts a message designed to feel normal.

BEC consistently ranks as one of the most financially damaging cybercrime categories. The FBI's 2023 Internet Crime Report listed BEC as responsible for over $2.9 billion in losses that year alone. Those numbers reflect actual wire transfers and data disclosures, not just attempted attacks.

BEC emails often pass traditional email security filters. They contain no malicious links, no malware attachments, and no obvious spam signals. They are plain text. They may even have passing SPF and DKIM records - because the attacker is spoofing a legitimate domain's From: header, not hacking the domain's actual mail infrastructure.

How AI Is Changing BEC at Scale

Three specific capabilities have shifted.

Automated reconnaissance and target selection. AI tools can scrape LinkedIn, company websites, and public filings to build a profile of potential victims within a target organization. They identify who handles finances, who has authority to approve payments, and who is likely to be out of the office during a holiday week. This used to take days of manual research. AI compresses it into minutes.

Personalized content generation at scale. A threat actor using traditional methods would write each BEC email by hand. AI language models can generate hundreds of unique, contextually appropriate messages in the time it takes a human to write one. The output is grammatically correct, professionally worded, and tailored to the recipient's role and department. Eye Security's analysis of one phishing kit found that it generated unique content for each recipient in a campaign, making detection based on content similarity nearly impossible.

Reply parsing and conversation takeover. AI tools can monitor incoming replies, extract relevant details - invoice numbers, payment amounts, bank account information - and generate appropriate responses without the attacker reading every message. One tool could maintain credible email conversations across dozens of simultaneous targets, automatically escalating serious opportunities to human handlers.

The result is that a single attacker or small team can run a BEC operation that previously required a team of writers, researchers, and coordinators.

Why Traditional Email Security Fails Against AI-Powered BEC

Content-based filters look for known malicious patterns: suspicious attachments, malicious URLs, spam keywords, known bad sender IPs. AI-generated BEC emails have none of those markers. They are well-written, targeted, and often contain no links at all - just a request to reply or call a number.

Reputation-based filters track sender behavior over time. A domain that suddenly starts sending high volumes of email gets flagged. But AI-powered BEC campaigns are designed to stay under the radar. Low volume, highly targeted, no malware - the signals that filters rely on are absent.

Sandbox detonation fails against pure-text BEC emails because there is nothing to detonate. The attack is in the social engineering, not the payload.

The common thread is that traditional email security inspects the message itself. AI-powered BEC succeeds by being a convincing message from a seemingly legitimate source. No amount of URL scanning or attachment sandboxing catches an email that contains only a request to call a number.

How DMARC Stops Domain Spoofing - The Core of BEC Attack Infrastructure

Most BEC campaigns rely on domain spoofing. The attacker sends email that appears to come from a trusted domain - your CEO's domain, your vendor's domain, your own domain - without actually being authorized to send from that domain.

Email authentication protocols exist specifically to prevent this.

SPF (Sender Policy Framework) checks whether the sending mail server's IP address is authorized by the domain's DNS records. If a server is not on the allowed list, SPF fails.

DKIM (DomainKeys Identified Mail) adds a cryptographic signature to email headers. Receiving servers verify this signature against a public key published in DNS. If the signature does not match, DKIM fails.

DMARC (Domain-based Message Authentication, Reporting and Conformance) sits on top of SPF and DKIM. It defines what should happen when an email fails authentication and provides reporting so domain owners can see who is trying to send email on their behalf.

Alignment is the critical concept. SPF and DKIM can both pass for a legitimate email, but DMARC requires that the domain used in the From: header aligns with the domain authorized by SPF or verified by DKIM. This is what stops spoofing: even if an attacker controls a mail server that can send from your domain, they cannot pass DMARC alignment without being authorized to use your domain in the From: header.

When you set your DMARC policy to p=reject, receiving mail servers refuse to deliver any email that fails DMARC alignment. The spoofed email never reaches the recipient's inbox. For organizations running DMARCFlow to monitor those alignment failures, the spoofing attempt appears in your reports - even when the email was blocked at the receiver.

What DMARC p=reject Actually Prevents - and What It Does Not

This distinction matters because it determines what kind of BEC DMARC can stop and what it cannot.

What DMARC p=reject prevents: Domain spoofing. When an attacker sends email that claims to be from your domain but has no authorization to use it, DMARC rejects the message. This covers BEC scenarios where the attacker spoofs your CEO's domain, your bank domain, or your vendor's domain to request a wire transfer or payment change.

What DMARC p=reject does not prevent: Lookalike domain attacks. An attacker registering "yourcompanye.com" (with an extra 'e') and sending from that domain passes DMARC because they own the domain they are sending from. The From: display name shows your company name, but the actual domain is different. DMARC has no mechanism to block this.

Display name impersonation is also outside DMARC's scope. An attacker sending from "attacker@gmail.com" but setting the display name to "Your CEO" passes DMARC because the domain in the From: header is gmail.com, which the attacker legitimately controls.

Understanding this scope is important for setting realistic expectations. DMARC is a strong defense against domain spoofing, which is the foundation of most sophisticated BEC campaigns. It removes one of the attacker's most reliable tools - but a complete defense strategy needs additional layers.

A Practical Defense-in-Depth Plan for AI-Era BEC

Deploy DMARC with p=reject once you have verified your legitimate email sources. Rushing to p=reject without knowing what sends email on your behalf causes legitimate email outages. Start with p=none or p=quarantine, read your DMARC reports, fix any unauthorized senders, then move to p=reject. DMARCFlow makes this practical by aggregating your reports and showing you exactly which sending sources are failing and why - before you change the policy.

Monitor your DMARC reports continuously, not just during policy transitions. Even with p=reject active, your reports tell you who is trying to spoof your domain and whether your own infrastructure is generating false failures. DMARCFlow automates this monitoring and alerts your team when suspicious patterns emerge - for example, a spike in authentication failures from an unusual country or ASN. When an AI-powered BEC campaign targets your domain, the attempted spoofing shows up in your reports before the attacker's emails reach any employee.

Train employees to verify unusual requests through a second channel. AI makes BEC emails more convincing, but the verification step remains effective. Any request for a wire transfer, payment change, or sensitive data should be confirmed via phone or a known-good internal communication channel - not by replying to the suspicious email.

Add secondary verification for vendor payment changes. A significant portion of BEC losses come from invoice fraud where the attacker impersonates a vendor and requests payment details be updated. Any payment detail change - regardless of how legitimate the email looks - should trigger a callback to a known phone number for that vendor.

FAQ

Does DMARC stop all BEC attacks?
No. DMARC with p=reject stops domain spoofing specifically. It does not stop lookalike domain attacks (attacker registers a similar domain name) or display name impersonation (attacker sets a misleading display name on a legitimate account). A complete BEC defense strategy includes DMARC, employee training, verification workflows, and monitoring for other attack vectors.

What does DMARC p=reject actually prevent?
It prevents unauthorized email from reaching inboxes. When a receiving mail server receives an email that fails DMARC alignment - meaning the From: domain does not match the domain authorized by SPF or verified by DKIM - and your policy is p=reject, the receiving server rejects the message. This blocks most domain spoofing attempts.

How does AI make BEC harder to detect?
AI automates the research, content generation, and reply handling that previously required human operators. Attackers can run more campaigns simultaneously, personalize each message to the recipient, and maintain convincing conversations without manual effort. The volume and personalization make traditional content-based and reputation-based filters less effective.

How do I move to DMARC p=reject safely?
Start with p=none to collect reports. Review who is sending email on your behalf and verify that all legitimate sources have valid SPF and DKIM. Fix any misconfigurations. Move to p=quarantine and monitor for two to four weeks. If no legitimate email is being quarantined, move to p=reject. Use DMARC reports throughout to catch any remaining unauthorized senders.