# Train, triage, repeat: AI-assisted phishing at scale and Red Canary's hybrid AI triage agent

> APWG observed over 3.8 million phishing attacks in 2025, with Q2 2025 (1,130,393) the highest quarterly total in two years, while adversaries use AI to craft convincing, well-formatted phishing emails at scale. Red Canary describes a hybrid triage agent (ML, rules engine, NLP feature extraction, LLMs via orchestrated subagents) reporting 94% classification accuracy.

- **Published:** 2026-10-26T12:44:34Z
- **Last reviewed:** 2026-10-26T12:44:34Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-4451
- **ID:** TL-2026-4451
- **Severity:** LOW
- **Category:** PHISHING
- **Status:** ACTIVE
- **Detections:** 9 · **IOCs:** 7 (full data via the Threadlinqs MCP server — Purple tier)

## Description

This is a trend and defensive-capability item rather than a single campaign. It names no CVE, malware family, threat actor or network IOC. Red Canary (author Ryan Morton) cites APWG data showing over 3.8 million phishing attacks in 2025, with Q2 2025 above 1.1 million, the highest quarterly total in two years. APWG's own counts are 1,003,924 in Q1 2025, 1,130,393 in Q2, 892,494 in Q3 and 853,244 in Q4, against 3.76 million in 2024. At this volume the article argues that no analyst team can triage unaided.

The article states that an AI tool can produce personal, well-formatted phishing emails that appear legitimate even to a trained analyst. It does not name specific tools, phishing kits or delivery mechanisms. The APWG 2025 Year in Review (published 2026-02-23) adds context: phishing and impersonation volume on social media grew between 51 and 843 percent quarter to quarter across major platforms. Finance, retail and federal government brands were most spoofed there, and social media and SaaS/webmail were the most attacked phishing categories. QR code phishing declined from Q3 to Q4 2025 as attackers moved to new infrastructure. Wire-transfer BEC attacks rose 136% in Q4 versus Q3, and APWG identified Scripted Sparrow as the most prolific BEC group, estimated at 6 million highly targeted emails per month. APWG also reported an average Q2 2025 wire-transfer BEC request of $83,099, up 97% from the prior quarter.

On the defensive side, Red Canary describes a graph workflow of five orchestrated subagents. (1) Parsing and enrichment normalizes the raw email into a standard object and adds domain reputation and abuse context. (2) Feature extraction produces true/false features through boolean logic and AI-powered NLP checks. (3) A rules engine applies deterministic, TTP-level rules before classification. (4) Hybrid classification combines a classical ML model trained on analyst assessments with LLM reasoning. (5) A transparency layer returns a category, summary and detailed explanation with every verdict. An analyst feedback loop retrains the system, and Red Canary reports 94% accuracy. That figure is a vendor claim and has not been independently verified. The publication date is not stated in the article body; a search summary reports 2026-06-30.

## MITRE ATT&CK

- T1566 Phishing
- T1566.002 Spearphishing Link
- T1566.003 Spearphishing via Service
- T1204.001 Malicious Link
- T1588.007 Artificial Intelligence
- T1583.001 Domains
- T1684.001 Impersonation

## Sources

- [Train, triage, repeat: The AI agent changing how we fight phishing (Red Canary, Ryan Morton)](https://redcanary.com/blog/threat-detection/phishing-ai-agent/)
- [Zscaler Blog mirror: phishing AI agent](https://www.zscaler.com/blogs/product-insights/phishing-ai-agent)
- [APWG Year in Review: Scams Explode in 2025 via Social Media, SMS, Email and QR Codes](https://www.newswire.com/view/content/apwg-year-in-review-scams-explode-via-social-media-sms-email-and-qr-22733100)
- [APWG Phishing Activity Trends Report, Q2 2025](https://docs.apwg.org/reports/apwg_trends_report_q2_2025.pdf)
- [APWG Phishing Activity Trends Report, Q4 2025](https://docs.apwg.org/reports/apwg_trends_report_q4_2025.pdf)
- [APWG Phishing Activity Trends Report, Q3 2025](https://docs.apwg.org/reports/apwg_trends_report_q3_2025.pdf)
- [APWG Phishing Activity Trends Report, Q1 2025](https://docs.apwg.org/reports/apwg_trends_report_q1_2025.pdf)
- [APWG Phishing Activity Trends Reports index](https://apwg.org/trendsreports)
- [Red Canary Builds Agentic AI Pipeline to Triage Phishing](https://letsdatascience.com/news/red-canary-builds-agentic-ai-pipeline-to-triage-phishing-8ab78a91)
- [Red Canary SecOps Weekly: How to defend against AI-powered attacks](https://redcanary.com/resources/videos/secops-weekly-how-to-defend-against-ai-powered-attacks/)
- [Red Canary SecOps Weekly: Email security and phishing](https://redcanary.com/resources/videos/secops-weekly-email-security-phishing/)

## Full data

Detection queries (Splunk SPL / Microsoft KQL / Sigma) and IOC values require the Threadlinqs MCP server (Purple tier): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-4451