Email Attack Taxonomy — StrongestLayer Threat Research 2026
StrongestLayer Threat Research · April 2026

The Evolution of Email Attacks

An interactive taxonomy of 37 attack subtypes across 6 categories, with detection likelihood ratings for SEGs, ML/Behavioral, LLM Wrapper, and Multi-Agentic Reasoning platforms.

37attack subtypes
4architectures compared
6threat categories
$55B
BEC Losses
68%
Below 0.30 Similarity
$4.45M
Avg Breach Cost
10K+
Orgs Hit by AiTM

Most email security stacks were built for a previous threat era

The attacks causing the most financial damage today contain no malicious payload, use no known-bad infrastructure, and look like normal business communication. BEC alone has produced $55 billion in cumulative losses since 2013. AI-generated spear phishing has collapsed personalization cost to near zero. AiTM phishing bypasses MFA by design.

These attacks share a common trait: they are structurally invisible to any detection system that only analyzes what arrives in the email itself. The gap is architectural, not configurational.

Key Finding

Traditional bulk phishing exhibits Jaccard similarity of 0.85–0.95. Advanced attacks from Q4 2025 showed average similarity of 0.458. 68% fell below 0.30 — where pattern-matching detection drops below statistical significance.

Four Eras of Email Attacks

2010–14
Volume Era
$5K–$25K/incident · Defender advantage
2015–18
Professionalization
$50K–$120K BEC · Shifting to attackers
2019–22
Infrastructure Era
$1.5M–$4M ransomware · Attacker advantage
2023+
AI Era
$4.45M avg breach · Zero-cost personalization

Four generations of email security

Each architecture represents a structural generation. Detection likelihood ratings are assessed against all four.

Generation 1
SEG
Operates on signals at delivery. Signatures, reputation, sandboxing.
Limit: Only sees what arrived
Generation 2
ML / Behavioral
Adds communication graph and behavioral baselines to flag anomalies.
Limit: No baseline for new/compromised senders
Gen 3 Component
LLM Wrapper
Applies LLM to email content for social engineering indicators.
Limit: Can only reason about given content
Gen 3 Platform
Multi-Agentic Reasoning
Actively retrieves missing context: follows links, detonates attachments, maps vendor infra.
Closes gaps by finding what the email doesn't reveal

37 attack subtypes across 6 categories

Click any cell to see the full breakdown: how it appears, evolution, and detection likelihood across all four architectures.

All
SEG Blind Spots
Highest Impact
AI-Era Threats

What to do next

1. Identify your current architecture

Determine which of the four architectural generations your primary email security solution represents. The detection ratings tell you what each architecture can and cannot do structurally, regardless of vendor or configuration.

2. Prioritize by your risk profile

Not every attack type is equally relevant. A mid-market company with significant vendor payment flows should prioritize Categories 1 and 3. Use the taxonomy grid above to identify where your current architecture rates Low or None.

3. Ask your current vendor these questions

For Credential Harvesting

Does your platform actively follow links to their final destination, including through redirect chains and AiTM proxies?

For BEC

How does your platform detect a fraudulent payment instruction sent from a legitimately compromised vendor account?

For Malware Delivery

How does your platform handle HTML smuggling, where the payload is assembled client-side and never transmitted as a file?

For Infrastructure Abuse

How does your platform evaluate email sent through a legitimate service like SharePoint or SendGrid that passes all auth checks?

4. Map the gap

For each attack type rated Low or None, assess business impact. Benchmarks: $50K–$120K for BEC, $1.5M–$4M for ransomware, $4.45M average breach cost where email was the initial vector.

Talk To Us

Your gateway can't see
what's already inside.

Deploy in minutes, not months. Zero tuning. See what your current tools are missing.