Book a demo →
← All posts
Threat Watch

TheMailer: A Phishing Kit That Writes a Different Email per Target

Varonis describes TheMailer, a self-hosted phishing kit that uses AI models to vary each recipient's subject, body and attachment and retest spam filters.

TheMailer: A Phishing Kit That Writes a Different Email per Target

AegisAI Threat Watch · AI-written phishing

TheMailer: A Phishing Kit That Writes a Different Email per Target

Varonis Threat Labs describes a self-hosted phishing kit, advertised since March 2026, that connects to AI models and varies the subject, body and attachment of each recipient's email. The Varonis post is undated; Dark Reading covered it in November 2026.

TL;DR
What
A self-hosted phishing kit that connects to AI models the buyer chooses and varies the subject, body and attachment of each recipient's email.
Who it hits
Not confirmed. A seller screenshot showed a target list of 145,133 entries, almost all Japanese business and university domains.
Why filters miss it
Each recipient can get a different version, so a bulk run leaves no repeated template to match, and the kit retests messages against spam filters.
Do today
Enforce DMARC at p=reject, check links through their full redirect chain at click time, and stop treating bad grammar as a warning sign.

Reported by Varonis Threat Labs. Dark Reading covered the research in November 2026.

TheMailer is a commercial phishing kit that hands most of a phishing campaign to AI models the buyer chooses. Varonis Threat Labs says it uses AI "at every stage of a phishing campaign", from drafting a lure to rewriting it until it passes a mail provider's spam filter.

Phishing-as-a-service is a model in which a seller rents or licenses a ready-made phishing toolkit to other criminals. Polymorphic phishing is phishing in which each message differs, so there is no fixed text, file or link to match. SPF, DKIM and DMARC are the email authentication standards that let a receiving server check whether a sending domain is genuine. Inbox Loop is the name Varonis reports for the kit's retest cycle.

Analysis based on research by Varonis Threat Labs. The post is undated; it says the kit has been advertised since March 2026. Dark Reading reported on it in November 2026. Varonis reports no victim count, no success rate and no indicators, and the screenshots it describes come from the seller. A related post covers another AI phishing platform that cloaks its pages from scanners.

24,829
emails in one campaign shown in a seller screenshot
145,133
entries in a target list the kit sorted by mail provider, per another screenshot
$3,000 to $6,000
advertised price of a lifetime license, depending on the deal and the AI modules
24,000+
views of the seller's sales thread, per Varonis

Source: Varonis Threat Labs. The screenshots come from the seller; Varonis says they are not a confirmed attack and reports no results.

Key judgments

  • The AI sits in the whole pipeline. Varonis describes AI steps for writing the email, rewriting it, retesting it and grouping recipients, and says earlier AI tools did less. We assess that a buyer needs little writing skill, since the buyer supplies an AI key, SMTP servers, proxies and a target list and the seller "handles the rest".
  • Every recipient can receive a different message. Varonis says templates can be regenerated and rotated so each victim gets a different version, subject lines are regenerated the same way, and PDF attachments are built so that no two files share a hash. We assess that a filter that groups near-identical messages has less to group.
  • Spam-filter feedback is built in. Varonis says the kit sends a test message, checks test inboxes over IMAP, and has an AI rewrite the template when a message lands in spam. The report adds that content is tuned for each mail provider.
  • The evidence is thin on results. Varonis says the kit "appears to have built a large customer base" and reports no success rate, victim count or customer number. It also describes no credential-capture page, only lure creation, delivery and redirect features.
  • The wording is polished but not researched. Dark Reading quotes a Varonis researcher describing the output as opportunistic phishing with the polish of something targeted, and says the kit has no features for researching individual targets.

Evaluating email security tools? The 2026 Buyer's Guide to Email Security has nine evaluation areas and the questions to ask vendors. Get the guide.

What does TheMailer automate for the person who buys it?

TheMailer is sold as a self-hosted platform on a lifetime license. Varonis reports an advertised price of $3,000 to $6,000, depending on the deal and whether AI modules are included, paid in cryptocurrency through an automated sales bot on Telegram, with free installation and support. The buyer brings an AI API key and a choice of model, SMTP servers, proxies and a target list.

Varonis names Claude, ChatGPT, Grok, Gemini, Mistral and DeepSeek as the supported models, connected through the operator's own API keys. Its report says the kit's configuration page notes that Anthropic and OpenAI block cloud IP ranges and suggests SOCKS5 or residential proxies. That describes what the tool tells operators; it does not show how any provider's controls performed.

What the kit does at each stage

StageFeature name reportedWhat Varonis says it does
Draft the lureAI Letter BuilderWrites an HTML email from a text description; real brand emails can be imported as templates
Rewrite the textAI Template GeneratorPresets include "Replace spam triggers" and "Remove AI patterns (humanize text, pass detectors)"
Vary each messageRegeneration and rotationEach recipient can receive a different body and subject line
Group recipientsAI AutopilotSorts targets by mailbox provider and applies sending limits and spam-filter notes for AI generation
RetestInbox LoopChecks test inboxes after a test send and has AI rewrite the template if it lands in spam
SendDirect-to-MX sendingSends over SOCKS5 proxies with randomized headers, content and sender profiles

Source: Varonis Threat Labs, from its description of the kit's panels and screenshots.

How does the kit make every email in a mass campaign different?

It changes the parts a filter would compare. Varonis says the body and the subject line are regenerated per recipient, and that HTML templates can be converted to PDF attachments with personalized filenames, such as an invoice named for the recipient. Embedded links and randomized metadata mean that no two files share a hash.

The same report says a File Templates tool builds SVG redirects and calendar invites, and that "trust redirects" hide malicious links behind Google domains. Varonis does not describe a credential-harvesting page or say where a victim lands. The sample subject it shows impersonates a brokerage with an "Unusual activity" alert.

The reports describe a seller screenshot of a campaign of 24,829 emails and another of an MX check on 145,133 target entries, almost all Japanese .co.jp and .ac.jp domains. Varonis presents these as screenshots from the sales thread. It does not report whether any message was delivered or opened.

How does TheMailer retest a message against spam filters?

The loop works on test inboxes, not on live recipients. Varonis says that after a test send the kit checks the test inboxes over IMAP, and if a message lands in spam the AI rewrites the template and the cycle repeats. Operators can steer the rewrite with an "AI focus" field, and AI Autopilot adds provider-specific spam-filter notes for Gmail, Microsoft, GMX, French providers and Yandex.

Varonis also reports tools that tune the sending side: Inbox Finder tests combinations of SMTP accounts, headers, encoding, DKIM settings and content against a seed mailbox, and an AI WarmUp feature that "appears to run" send-and-reply exchanges between new accounts. Varonis hedges the last one, and so does this article.

How does TheMailer compare with Bluekit?

BleepingComputer reported on 30 April 2026 on Bluekit, a phishing kit with an AI Assistant panel that supports Llama, GPT-4.1, Claude, Gemini and DeepSeek to help draft phishing emails. The kit also offers more than 40 templates and combines domain purchase, phishing page setup and campaign management in one panel. Varonis presents TheMailer as going further by building AI into how the whole platform works. Neither report offers a feature-by-feature comparison.

Why do template matching and file hashes miss a kit like TheMailer?

This section is AegisAI's analysis. A control that clusters near-identical messages, blocks a known file hash or looks for a fixed phrase needs a repeat to act on. The features Varonis describes remove the repeat, and the retest loop removes the messages that a provider's own filter would have caught.

Where each control stands

Six controls are compared against the features Varonis reports. The research tests none of them.

ControlWhat it sees in this campaignResult
Content and template matchingA different subject and body for each recipientHas little to match; rewrites are tested against provider filters first
File hash and attachment blocklistsPDFs built so that no two share a hash, with personalized filenamesA known-bad hash does not recur
Link scanning at deliveryEmbedded links and redirects behind trusted domainsDepends on following the full redirect chain at the moment of the click, as Varonis advises
SPF, DKIM and DMARC at p=rejectMail sent from the attacker's own infrastructure and sender profilesStops spoofing of the organization's own domain; says nothing about a domain the attacker owns
Training on grammar and typosFluent text with AI-writing patterns removedBad grammar is no longer a reliable signal
Review of the request and the senderAn urgent request or an "unusual activity" alert from an unfamiliar senderApplies when the check reads what the message asks for and who it claims to be from

AegisAI's reading of the reported features. Varonis does not test these controls.

When every email in a campaign is different, a repeated template is no longer a reliable signal, and what remains to judge is the request, the sender and the destination.

AegisAI Threat Intelligence

What this means for a security team

This section is AegisAI's analysis. Exposure depends on three things: whether detection relies on repetition across messages, whether links are checked when they are clicked and not only when mail arrives, and whether staff training still treats poor writing as the main sign. Varonis frames the output as opportunistic, so the realistic expectation is large volumes of plausible, generic lures, not a hand-built message for each person.

Varonis reports no confirmed attack and the target list it shows is mostly Japanese domains. The same features would work against any organization whose staff receive email, but the research does not say that happened.

Why standard controls miss it

This section is AegisAI's analysis. The signals that standard controls lean on, repeated text, repeated files and bad writing, are the ones the kit varies or removes. A message that is new, fluent and sent from fresh infrastructure gives a content filter nothing to compare it with.

How AegisAI protects against it

AegisAI has not observed this kit or any campaign sent with it. Varonis describes TheMailer as an email platform, so the approach below applies to messages delivered to a work mailbox.

AegisAI's agents evaluate the message and the destination it points to together, and pull a malicious message out of mailboxes before users see or click it. More on how that works: inbound email security.

When a lure or a compromised session is caught, the response compounds: the message is withheld at classification, copies already delivered are retracted, rewritten links refuse a late visit, the account's outbound and internal mail gets rescanned, and lateral phishing attempts to colleagues are blocked at the same bar rather than treated as a fresh, unrelated message.

How AegisAI stops credential phishing, 48 second video

How AegisAI stops credential phishing, 48 seconds

Recommendations for defenders

  • Enforce SPF, DKIM and DMARC at p=reject on the organization's own domains, and watch for lookalike and newly registered domains, as Varonis recommends. This limits spoofing of the organization; it does not stop mail from a domain the attacker owns.
  • Check links through their full redirect chain at click time, including links behind trusted domains, and scan the destination, not only the message text.
  • Update training so that bad grammar is not treated as a tell. Teach staff to question urgency, unexpected requests, login prompts and "unusual activity" alerts, and give them one place to report them.
  • Add detections that do not need a repeated template: sender and domain age, mismatched reply paths, PDF attachments whose links lead to a sign-in page, and requests that do not fit the sender's normal pattern.
  • Apply least-privilege access and data monitoring, so that a successful click reaches as little as possible.

FAQ

What is the TheMailer phishing kit?

TheMailer is a commercial, self-hosted phishing kit, advertised since March 2026, that connects to AI models such as Claude, ChatGPT, Grok, Gemini, Mistral and DeepSeek through the buyer's own API keys. Varonis says it uses AI at each stage of a campaign, from writing the lure to retesting it against spam filters.

How do I stop mass AI-generated phishing campaigns that change every email?

Stop relying on content matching alone. Enforce SPF, DKIM and DMARC, watch for lookalike and newly registered domains, check links through their full redirect chain at click time, and train staff to question urgency and unexpected requests instead of looking for typos. Add detection that judges the request, the sender and the destination. These steps combine Varonis's advice with AegisAI's analysis.

How much does TheMailer cost?

Varonis reports an advertised lifetime license of $3,000 to $6,000, depending on the deal and whether AI modules are included, paid in cryptocurrency. The buyer also supplies an AI API key, SMTP servers, proxies and a target list.

Is TheMailer a spear phishing tool?

Not according to Dark Reading, which quotes Varonis researcher Daniel Kelley describing the personalization as closer to a name on an invoice than research into a victim. He called the output "opportunistic phishing with the polish of something targeted".

For researchers

What the reporting does not tell us 5 gaps

Results. Varonis reports no success rate, no victim count, no customer count and no confirmed attack. The large customer base is stated with "appears".

The screenshots. The campaign of 24,829 emails and the 145,133-entry target list come from seller screenshots. Varonis does not say whether any message was delivered.

Dates. The Varonis post is undated. It says the kit has been advertised since March 2026 and that the research is recent.

Provider controls. Varonis reports the kit's own notes about cloud IP ranges and proxies. It does not test whether any AI provider's controls were bypassed.

Indicators and scope. No domains, IP addresses, hashes or sample emails are published. This article rests on the Varonis post; the Kelley quotes come from Dark Reading. AegisAI did not test or observe the kit.

Reported identifiers 4 items

The research publishes names and patterns, not network indicators. Treat these as labels for searching, not a blocklist.

TypeIdentifierRole
Kit nameTheMailerSelf-hosted AI phishing platform sold on a cybercrime network
Feature namesAI Letter Builder, AI Template Generator, AI Autopilot, Inbox LoopPanels Varonis names for drafting, rewriting, grouping and retesting
Attachment patternInvoice_%name%.pdfExample of a personalized PDF filename from the kit's templates
Sales channelTelegram channel and automated botWhere the license is sold, with payment in cryptocurrency
MITRE ATT&CK mapping 3 techniques

These techniques are AegisAI's mapping of the reported behaviour. Varonis does not map the kit to technique IDs. IDs and names were checked against attack.mitre.org on 19 November 2026.

IDNameUse in this campaign
T1588.007Obtain Capabilities: Artificial IntelligenceThe kit uses commercial AI models through the buyer's API keys to write and rewrite lures.
T1566.001Phishing: Spearphishing AttachmentTemplates are converted to PDF attachments with personalized filenames and unique hashes.
T1566.002Phishing: Spearphishing LinkEmails carry embedded links and redirects, including trust redirects behind Google domains.
References 4 sources

Phishing that reaches the inbox, stopped before anyone clicks.

Book a demo