High-value BEC persistence, AI-scaled approval phishing, and deepfake verification fraud
Key Questions
What are the latest trends in BEC and vendor email compromise attacks?
BEC accounts for 61-74% of incidents with heavy use of AI for payroll spoofing, invoices, and deepfakes, leading to losses like $25M at Wells Fargo and $25.6M at Arup; FBI reports exceed $3B annually. VEC often employs lookalike domains that bypass DMARC when attackers control the domain and set permissive policies.
How do attackers achieve long-term persistence in high-value email compromises?
One case showed a global stock exchange executive compromised for five months using legitimate tools like Aspose and Dropbox for exfiltration with repeated inbox theft. Techniques include inbox rule manipulation, shared access links, session hijacking via MFA bypass, and living-off-the-land approaches that avoid malicious attachments or URLs.
What makes deepfake-based BEC particularly challenging to defend against?
Deepfake fraud reached $3.7B with 89% occurring in 2025-2026, and detection tools lose 45-50% accuracy in real conditions while attacks can originate from social media or use video conference impersonation. Durable defenses focus on process changes like out-of-band verification and dual approvals rather than relying solely on detection tools.
How are attackers combining evasion tactics in recent BEC phishing campaigns?
Campaigns use link laundering through trusted security vendor wrappers, callback phishing via Google Calendar without malicious links, and compromised nonprofit domains mixing clean and malicious content to bypass SPF/DKIM/DMARC. Additional methods include Lua loaders disguised as font files and DocuSign-themed kits delivering RMM tools like ScreenConnect.
What controls are recommended to counter AI-powered BEC and approval phishing?
Experts advocate deterministic controls, supplier DMARC enforcement with RejectDirectSend policies, CASB/DLP for data exfiltration, and relationship history checks over manual review. ARToken-style BEC-as-a-service platforms and dark web deepfake tools highlight the need for layered defenses as AI lowers barriers for attackers.
AI-assisted campaigns are scaling reconnaissance, vendor impersonation, thread injection, synthetic invoices, executive impersonation, and deepfake voice/video approvals. AI agents add risk when delegated authority, prompt injection, or integrations enable payment actions without human review; independent verification, least privilege, explicit approval, dual control, and transaction monitoring remain more dependable than sender inspection or synthetic-media detection alone. Recent vendor-led training coverage emphasizes multichannel and deepfake simulations but provides limited independent evidence.