← Tuesday, November 3 deck

17Security1 source posting

A User Says ChatGPT Sent His Wife to a Near Perfect Lookalike of the T3 Code Download Site

ChatGPT just tried to get my wife to download malware or something 💀 asked for the T3 Code download → got t.co/07oJHAdy96 almost pixel-perfect clone, registered 2 weeks ago real one is t.co/rbsSUGZNJt AI search is the new phishing vector @theo

Screenshot of a ChatGPT answer recommending the official T3 Code macOS app, with a preview card for www.t33.codes showing a T3 Code page that reads The open source control plane for coding agents and a Download T3 Code title.

Context

Why it matters

If accurate, it shows how AI search can surface a lookalike site as the official download, turning chat answers into a new route for phishing and malware.

T3 Code is a control panel for coding agents made by Theo Browne's company T3. Its real site is t3.codes. Phishing sites often copy a real product's look and use a domain that differs by one character to trick people into downloading malware.

Deep dive

Dorian D. (@doriandeacu) posted on Nov 3 ET that ChatGPT tried to get his wife to download what he suspects was malware.

According to his post, she asked ChatGPT for the T3 Code download and got a link to t33.codes, which he describes as an almost pixel perfect clone registered two weeks earlier. He says the real site is t3.codes. He tagged Theo Browne (@theo), whose company makes T3 Code, and wrote that AI search is the new phishing vector.

His screenshot shows a ChatGPT answer recommending the official macOS app with an Official download link, and a preview card for www.t33.codes that copies the T3 Code page design.

This is one user's report. The post does not confirm what the lookalike site actually delivered, and OpenAI had not commented in the posts found during this scan. It is not clear whether the result came from ChatGPT's web search ranking a fake site or from another cause.

The post passed 100,000 views. Lookalike domains are a long standing phishing tactic, but AI assistants that present a single confident link can make them harder to spot than a list of search results.

What to expect next

Watch for whether OpenAI responds or blocks the lookalike domain in ChatGPT search results. Security researchers may examine what the site serves. The T3 team may warn users directly. The case adds to pressure on AI search products to verify official download links and flag newly registered domains, a problem that will grow as more people get software links from chatbots.