# TheMailer phishing kit writes a different email for every target

Researchers say the self-hosted kit uses the buyer's own AI model to rewrite lures until they reach the inbox, so bulk phishing no longer looks like bulk phishing.

Source: https://junglewise.ai/news/themailer-phishing-kit-writes-a-different-email-for-every-target  
Byline: Junglewise News Desk, https://junglewise.ai/news  
Topic: Threat campaigns  
Published: 19 November 2026. Updated: 19 November 2026  
Reading time: 6 min

![A sealed vintage white envelope, back side showing the triangular flap seam, on a black background](https://images.unsplash.com/photo-1649019489428-70f505daacd6?ixid=M3w0NDU4NjF8MHwxfHNlYXJjaHwxfHxlbWFpbCUyMGVudmVsb3BlfGVufDF8MHx8fDE3OTUwNDY4MzF8MA&ixlib=rb-4.1.0&w=1600&q=80&fm=jpg&fit=crop)

Picture: Photo by [Valeria Reverdo](https://unsplash.com/@lereverdo?utm_source=junglewise&utm_medium=referral) on [Unsplash](https://unsplash.com/?utm_source=junglewise&utm_medium=referral)

A commercial phishing kit called TheMailer gives criminals a way to send tens of thousands of emails that are each worded differently, which undermines one of the oldest defences in email security: spotting a mass campaign because every message looks the same. The kit also tests its own output, sending trial messages and checking over IMAP whether they landed in the inbox or the spam folder, then rewriting them until they get through.

Phishing emails built with this kit are tailored to the recipient's mail provider and adjusted specifically to defeat content filtering, and screenshots from the seller show the scale involved: a target list of 145,133 contacts, almost all of them Japanese `.co.jp` and `.ac.jp` domains belonging to businesses and universities.

The research describing the kit was published by Varonis and written up by senior security researcher Daniel Kelley, and was [reported by Dark Reading](https://darkreading.com/perimeter/themailer-phishing-kit-ai-drivers-seat) and summarised by [SC Media](https://scworld.com/brief/new-ai-powered-phishing-kit-automates-malicious-email-campaigns). Most criminal tools that use AI so far have used it for one job at a time: writing a lure, coding a fake login page, doing reconnaissance, or automating small repetitive tasks. What stands out about TheMailer, according to the research, is that it hands most of the phishing workflow to an AI model the buyer chooses and pays for.

That includes tailoring the wording to each mail provider and, as Kelley put it, the ability to "rewrite them until they bypass spam filters". The software is self-hosted and sold through a Telegram channel with an automated cryptocurrency sales bot, and the seller throws in free installation and support. The buyer supplies SMTP servers (the servers that actually send mail), proxies and a list of target addresses, and the kit handles the rest.

Kelley is careful about how he frames the threat. This is not a tool for researching named executives, he says: personalisation goes about as far as putting the right name on an invoice, and Varonis saw no features for researching individual targets. "It's opportunistic phishing with the polish of something targeted," Kelley told Dark Reading.

## What to do now

None of this needs a new product. It needs the basics done properly, because the thing that used to make bulk phishing obvious, identical wording across hundreds of messages, is gone.

1. **Get email authentication in place and enforced.** Varonis recommends SPF, DKIM and DMARC enforcement. These are DNS records that let other mail servers check that mail claiming to be from your domain really is, and they make it harder for someone to spoof your organisation to staff, customers or suppliers. If your DMARC policy is still set to "none", move it towards quarantine or reject once you have confirmed your legitimate senders pass.

2. **Rewrite phishing training for a world without typos.** Varonis is blunt on this point: "Bad grammar isn't a reliable sign anymore, so teach users to be wary of urgency, unexpected requests, and login prompts". SC Media made the same point, suggesting staff be trained to recognise psychological triggers such as manufactured urgency rather than relying on outdated tells. Make sure your simulated exercises reflect that, not the clumsy examples of five years ago.

3. **Check links at click time, through the whole redirect chain.** Varonis recommends following redirects all the way to the final destination rather than judging the link that appears in the message. A link that looks harmless in an email can pass through several hops before it lands on a credential-harvesting page.

4. **Watch for lookalike and newly registered domains.** Domains that resemble yours, your bank's or a major supplier's are a standing warning sign, and Varonis recommends monitoring for them.

5. **Assume someone will click, and limit what happens next.** Varonis recommends least-privilege access and data monitoring to contain the damage when a user clicks anyway. In practice that usually means multi-factor authentication on email and remote access, plus cutting back accounts that can reach far more data than the job requires. If credentials are phished, the damage depends on what those credentials could open.

6. **Shift what you look for.** Because each recipient gets a uniquely rewritten message, there is no single template for a content filter to block, so detection has to focus on intent and behaviour, [SC Media reported](https://scworld.com/brief/new-ai-powered-phishing-kit-automates-malicious-email-campaigns). That means paying attention to unusual sender relationships, odd login locations and sudden mailbox rule changes rather than to the wording of the email.

## Technical details

TheMailer is sold as a lifetime licence for between $3,000 and $6,000 in cryptocurrency, with the price depending on the deal and which AI modules are included, according to Varonis's research as reported by Dark Reading. The sales thread has been viewed more than 24,000 times. Distribution runs through a Telegram channel and an automated crypto sales bot, with installation and support included at no extra cost.

The buyer plugs in their own API key and picks a model. Supported options named in the research include Claude, ChatGPT, DeepSeek, Grok, Gemini, Mistral and the OpenRouter aggregation service. That design matters for defenders: the operator does not host a model, they rent capability from a mainstream provider and point the kit at it.

Functionally, the kit can build a lure from a short prompt or from a real brand email, personalise the recipient address, body text and subject line for each target, strip out the tell-tale patterns of AI writing, and adjust content specifically to get past spam filtering. Kelley's description of the effect is that "every recipient gets their own version of the email, subject line, and attachment, so a bulk campaign no longer looks like one".

The component Varonis calls Inbox Loop is the part that closes the circle. It sends test messages, connects over IMAP to see whether they were filtered as spam, and uses the configured AI model to rewrite and resend until delivery succeeds. Spam filtering, in other words, becomes a feedback signal rather than an obstacle.

On scale, seller screenshots show a single campaign of close to 25,000 emails and an Autopilot feature sorting 145,133 contacts by mail provider. That list is overwhelmingly Japanese business and university domains. The figures come from the seller's own promotional material, so they are claims rather than measured results.

What is not known is how well the kit actually works. Varonis says it is unclear how successful TheMailer is in practice, while assessing that it has a large customer base. There is no public victim count, no named campaign and no indicators of compromise in the research as reported.

Both Kelley and researchers cited by SC Media expect the capability to spread. Kelley's view is that once one seller offers this, buyers will start asking other sellers for it, and SC Media reported that researchers expect automation on this scale to become a standard feature of criminal tooling. Varonis, which published the research, also [sells an email security product](https://bleepingcomputer.com/news/security/meet-varonis-interceptor-ai-native-email-security) aimed at this class of attack, which is worth noting when weighing the recommendations.

## What changed

- 19 November 2026: Added new details: OpenRouter support, Telegram sales with a crypto bot, the IMAP-based Inbox Loop deliverability tester, and a target list of mostly Japanese domains.

## Related articles

- [Attackers are running AI agent frameworks that turn patches into exploits](https://junglewise.ai/news/attackers-are-running-ai-agent-frameworks-that-turn-patches-into-exploits)
- [Microsoft signing credential suspected in poisoned agent-framework PyPI releases](https://junglewise.ai/news/microsoft-agent-framework-python-package-briefly-poisoned-with-a-credential)
- [Phishing emails posing as Microsoft 365 outage notices ask staff to restore held mail](https://junglewise.ai/news/phishing-emails-posing-as-microsoft-365-outage-notices-ask-staff-to-restore-held)
