TheMailer Uses AI to Rewrite Phishing Emails Until They Reach Inboxes

Summary
Varonis Threat Labs describes TheMailer, a self-hosted phishing platform that uses commercial AI models to personalize lures, tailor emails to mailbox providers, and rewrite messages that land in spam.
Key points
- The self-hosted platform uses operator-supplied API keys to connect to commercial AI models.
- Its Inbox Loop checks test messages via IMAP and uses AI to rewrite messages that land in spam, then retries.
- AI tools generate and personalize email templates and subject lines, and adapt content to different mailbox providers.
- The platform supports sending through proxies and offers features for testing sender settings and managing replies.
- It can create personalized PDF attachments and SVG or calendar lures; the article describes examples impersonating financial and account-security messages.
- The article recommends enforcing SPF, DKIM, and DMARC, checking links through full redirect chains, monitoring sender behavior, and training users not to rely on poor grammar as a phishing clue.
Article Details
- Attack Vectors
- TheMailer generates and sends personalized phishing emails, including branded account-security and financial-service lures, and rewrites messages that land in spam to improve inbox delivery.
- The platform creates personalized PDF attachments with embedded links, as well as SVG redirects and calendar invites.
- Operators can send directly to MX servers using SOCKS5 proxies and use randomized sender profiles, headers, and content.
- The article reports targeting of Japanese business and university addresses, and lures aimed at financial services and hotel bookings.
- Defensive Notes
- Enforce SPF, DKIM, and DMARC with a reject policy.
- Monitor for lookalike and newly registered domains, and inspect links through the full redirect chain at click time, including links hidden behind trusted domains.
- Train users to watch for urgency, unexpected requests, and login prompts; do not rely on poor grammar as a phishing signal.
- Use least-privilege access and data monitoring to limit harm if a user clicks a phishing link.
MITRE ATT&CK
T1090.002 · External ProxyOperators are instructed to route connections through SOCKS5 or residential proxies, including to bypass AI providers’ restrictions on cloud IP ranges.T1566.001 · Spearphishing AttachmentTheMailer generates personalized PDF attachments and other lures, including calendar invites, for phishing campaigns.T1566.002 · Spearphishing LinkTheMailer creates phishing emails with embedded links and builds SVG redirects; the article also describes links hidden behind trusted domains.T1656 · ImpersonationTheMailer can create lures impersonating brands, including a sample Robinhood unusual-activity alert and a Robinhood-branded SVG.
Vendors
AnthropicChatGPT, Grok, Gemini, Mistral, DeepSeek, and OpenRouter. The tool’s configuration page notes that Anthropic and OpenAI block requests from cloud IP ranges, and instructs operators to route traffic through SOCKS5OpenAIGrok, Gemini, Mistral, DeepSeek, and OpenRouter. The tool’s configuration page notes that Anthropic and OpenAI block requests from cloud IP ranges, and instructs operators to route traffic through SOCKS5 or
Products
ChatGPTrecently identified a self-hosted phishing platform that uses leading AI services, including Claude, ChatGPT, Grok, Gemini, Mistral, and DeepSeek, at every stage of a phishing campaign. Operators simply plug inClaudeThreat Labs recently identified a self-hosted phishing platform that uses leading AI services, including Claude, ChatGPT, Grok, Gemini, Mistral, and DeepSeek, at every stage of a phishing campaign. Operators simplyDeepSeekphishing platform that uses leading AI services, including Claude, ChatGPT, Grok, Gemini, Mistral, and DeepSeek, at every stage of a phishing campaign. Operators simply plug in their own API key and pick the model.Geminia self-hosted phishing platform that uses leading AI services, including Claude, ChatGPT, Grok, Gemini, Mistral, and DeepSeek, at every stage of a phishing campaign. Operators simply plug in their own API keyGrokidentified a self-hosted phishing platform that uses leading AI services, including Claude, ChatGPT, Grok, Gemini, Mistral, and DeepSeek, at every stage of a phishing campaign. Operators simply plug in their ownMistrala self-hosted phishing platform that uses leading AI services, including Claude, ChatGPT, Grok, Gemini, Mistral, and DeepSeek, at every stage of a phishing campaign. Operators simply plug in their own API key and pickOpenRouterthrough operator-supplied API keys, including Claude, ChatGPT, Grok, Gemini, Mistral, DeepSeek, and OpenRouter. The tool’s configuration page notes that Anthropic and OpenAI block requests from cloud IP ranges, and
Tools
Countries
FranceIn the AI Autopilot module, recipients are grouped by mailbox provider, including Gmail, Microsoft, GMX, French ISPs, and Yandex. The module includes sending limits and “spam-filter notes that guide AI generation,” soGermanyIn one screenshot, a visible sender is “Kundenservice IONOS,” the German customer service address of hosting provider IONOS.Japancontent.One screenshot shows Autopilot sorting a list of 145,133 entries by provider, almost all of them Japanese .co.jp and .ac.jp domains belonging to businesses and universities.Russiaservices and account security lures, hotel booking callback lures, and targeting of German, French, and Russian-language mailbox providers.
