Threat intel5 min read
TheMailer: a phishing kit that lets AI rewrite emails until they beat spam filters
A self-hosted phishing kit sold since March 2026 plugs Claude, ChatGPT and other AI models into lure writing, delivery testing and reply handling, rewriting messages that land in spam.
4Tify Team

A phishing kit called TheMailer is being sold on a cybercrime forum with commercial AI models wired into almost every step of an email campaign. Its standout trick is simple: when a test message lands in spam, the kit asks an AI model to rewrite it and sends again, until the message reaches the inbox.
What happened
TheMailer (marketed as TheMailerPRO) is a self-hosted bulk mailer built for phishing. It has been advertised on an underground forum since March 2026, where its sales thread has passed 24,000 views. That number shows interest, not buyers. Lifetime licences cost between $3,000 and $6,000, with the higher price including the AI modules. Sales run through a Telegram channel and an automated cryptocurrency payment bot, and buyers get installation help.
Buyers bring their own infrastructure: mail servers, proxies, target lists and API keys. The kit connects to Claude, ChatGPT, Grok, Gemini, Mistral and DeepSeek through whichever keys the operator supplies.
How it works
The feature that matters most is called Inbox Loop. The kit sends a test email to a mailbox it controls, logs in over IMAP and checks whether the message arrived in the inbox or the spam folder. If it went to spam, the template goes back to the AI model for a rewrite, and the cycle repeats. Content is tuned against real filter results rather than written once and hoped for.
The other modules fill in the rest of a campaign:
- AI Letter Builder produces HTML emails from a short prompt and can import genuine messages from a connected mailbox to reuse as templates.
- Per-recipient rewriting changes subjects and bodies for each target. Presets include account security alerts and routine transaction notices, plus an option meant to strip "AI patterns" from the wording.
- AI Autopilot sorts recipients by mailbox provider (Gmail, Microsoft, GMX, Yandex and French providers) and applies sending limits per provider.
- Warm-up and testing tools exchange messages between sending accounts so they look established, and test sender, header and content combinations against seed mailboxes.
- A reply manager collects answers from targets, with AI sorting and drafting responses.
Attachments get the same treatment. HTML templates become PDFs with the recipient's name in the filename, and randomised metadata means each file has a different hash. The kit also builds SVG files that redirect to phishing pages, and calendar invitations. Sample lures carry Robinhood and IONOS branding, along with financial alerts and hotel booking themes, and links can sit behind Google redirect domains so the first hop looks trusted.
Who is affected
Anyone whose staff read email. Screenshots from the kit show a target list of 145,133 addresses, most at Japanese company and university domains, and a seller screenshot shows a campaign of 24,829 emails. Neither proves that any organisation was compromised. There is no public data yet on how many people bought the kit or how well its campaigns perform.
What it does show is that filter evasion and lure writing, once the hard parts of running phishing at scale, are now packaged behind a dashboard that looks like ordinary marketing software.
What to do now
Spotting clumsy grammar no longer works as a filter for people or tools. Focus on identity, intent and behaviour:
- Enforce email authentication. Publish SPF and DKIM for every domain you own, including ones that never send mail, and move DMARC to
p=reject. The NCSC guidance on email security and anti-spoofing walks through the setup. - Check links at click time. Follow the full redirect chain, including links that start on a well-known domain. A trusted first hop says nothing about the final page.
- Don't trust file hashes alone. Personalised PDFs and SVGs defeat simple blocklists, so inspect attachment behaviour and treat SVG files from outside senders with suspicion.
- Watch for lookalike and newly registered domains that imitate your brand or your suppliers.
- Train for the request, not the writing. Unexpected requests for passwords, payments or verification codes should be checked through a known contact channel, however natural the message sounds. CISA's phishing guidance covers both the technical and user-side controls.
Kits like this are bought, discussed and supported on the same forums where stolen credentials later turn up. Our Deep/Dark Web monitoring service watches those spaces for mentions of your domains and for leaked logins, so a successful lure is caught early instead of after the account is abused.