AI-powered phishing attacks bypass email filters
Attackers now use AI to research targets, write convincing phishing emails, and even fake video calls with deepfake colleagues to trick employees into approving fraudulent payments. Traditional email filters that only scan for bad links or attachments cannot catch these because there is no malicious file to detect, only a convincing fake conversation.
- Report priority
- Medium
How it works
An AI agent scrapes a company's public website, job posts, GitHub pages, and executive social media to build a personalized, believable pretext in seconds, then sends that tailored lure by email and can escalate to a deepfake voice or video call if the target hesitates.
What to do
Most readers are not the target of any single campaign, but anyone who receives an urgent payment request, executive impersonation, or unexpected video call asking for a wire transfer or sensitive action should treat it as a possible AI-crafted lure, especially if it plays out over more than one channel.
Verify unusual payment or verification requests through a separate, previously known contact method rather than the one the request arrived, and treat surprise video or voice calls involving money movement with the same suspicion as a phishing email, even if the faces and voices look and sound familiar.
Technical details
At the engineering firm Arup, an attacker opened with a phishing email that impersonated the company's UK-based CFO. When the employee hesitated, the attacker followed up with a deepfake video call showing what looked like several familiar colleagues, but every other face on the call was synthetic. The employee ended up approving 15 transfers worth about $25 million.
The piece traces phishing's evolution from malicious payloads (links, attachments) that secure email gateways can scan for, to pure social-engineering business email compromise that gateways cannot flag because there is no bad content to detect, to a third stage where generative AI and deepfakes handle reconnaissance, drafting, sending, and adapting the lure across email, collaboration tools, and live calls without much human involvement on the attacker's side. Microsoft has tracked phishing platforms generating tens of millions of messages a month, and a 2026 Dark Reading poll found 48% of security professionals rank agentic AI as the top attack vector for the year.