Email is one of the most sensitive documents archives most people maintain. Bank statements, medical correspondence, legal notices, tax forms, contracts, personal letters — an inbox accumulated over years contains a detailed, timestamped record of a person’s financial, medical, legal, and personal life.
And now AI email clients can summarize all of it.
Third-party AI email applications promise to help you manage inbox overload: summarize long threads, surface important messages, draft replies, search your entire email history by meaning rather than keyword. These features are genuinely useful. They also require the application to read — and in most cases, transmit to an AI processing service — the full content of your emails, including attachments.
Most users using these tools have not read the privacy policies of the AI services processing their inbox.
How AI Email Clients Work
The major players in the AI email client space — apps like Superhuman, Shortwave, Spike, Mimestream, and several newer entrants — connect to your existing email account (Gmail, Outlook, and similar) via standard API access. You grant the app permission to read your email, and it becomes an interface layer between you and your inbox.
The AI features work by sending email content — messages, threads, and attachments — to an AI processing layer. Depending on the app, this might be a proprietary model the company runs internally, or it might be a third-party AI API (commonly OpenAI’s API, Anthropic’s API, or Google’s Vertex AI). The content is processed, and the result (summary, draft reply, search result) comes back to your interface.
At each step, the content of your emails and attachments is transmitted and processed by systems beyond your email provider. The email provider (Google, Microsoft) has already seen the content — that’s not new. The AI email client, and the AI service it uses, are additional parties who now process the same content.
What “Attachments” Actually Means
Email attachments are not a uniform category. The phrase calls to mind PDFs and spreadsheets, but in practice email attachments include:
- Bank statements and credit card bills
- Tax documents (W-2s, 1099s, tax returns)
- Medical records, lab results, and health insurance documents
- Legal correspondence, contracts, and court documents
- Employment documents including offer letters, performance reviews, and severance agreements
- Immigration documents, visa applications, and passport scans
- Real estate documents including leases, mortgage statements, and property records
- Personal photographs
- Confidential professional files forwarded from work accounts
When an AI email client promises to “search across your entire inbox,” it means it can surface content from all of these documents in response to a query. Achieving that requires the system to have indexed — which means read and processed — all of them.
What the Privacy Policies Actually Say
Privacy policy language in this space follows familiar patterns, but the specific formulations matter.
Sub-processors. When an AI email app uses a third-party AI API to process your content, the third-party AI provider becomes a sub-processor of your data. Responsible apps list these sub-processors explicitly and provide links to their privacy policies. The chain of data flow is: your email → AI email app servers → AI processing service (OpenAI, Anthropic, Google) → back to you. The AI processing service processes your content under their data processing terms, which may differ materially from the email app’s own terms.
Training use. The critical question is whether your email content or attachment content is used to train AI models. Enterprise tiers of AI APIs typically prohibit training on customer data, and some email apps claim this protection explicitly. Consumer tiers often work differently. If an AI email app is using a consumer-tier AI API (or if the terms are silent on this point), your email content may be used to train the underlying model.
Retention. Most AI email apps do not permanently store your email content — they process it in real time to generate summaries and search indexes, and the underlying content lives in your email provider. But the search indexes, summaries, and processed metadata they generate from your emails may be stored, and the retention schedule for that derived data is often different from the email content itself.
Data sovereignty. Where your email content is processed matters for legal purposes. If the AI processing service operates under US law, your emails may be accessible to US legal process even if you’re in the EU and your email provider maintains EU servers.
The Third-Party Attachment Problem
A specific scenario worth understanding: you receive a document via email from a third party — a doctor, a lawyer, an accountant, a bank. That party sent it to you for a specific purpose, in a specific context. They did not consent to that document being processed by an AI email application’s machine learning infrastructure.
This isn’t a theoretical concern. It’s a practical privacy and, in some cases, professional ethics issue. Attorneys in many jurisdictions have ethical obligations regarding client information. Medical professionals have HIPAA obligations regarding patient data. When someone sends you a document under these professional obligations, they generally have not considered that you might use an AI email client that processes attachments through a third-party AI API.
The question of whether you violate another party’s privacy (or their professional obligations) by using an AI email client to process attachments sent to you is not settled. But it’s worth being aware of when the attachment in question is a medical record, a legal document, or other professional correspondence.
Gmail vs. Third-Party AI Clients: The Privacy Difference
It’s worth distinguishing between Google’s own AI features in Gmail and third-party AI email clients.
Google’s integration of Gemini into Gmail operates within Google’s existing privacy commitments for Gmail. The processing happens within Google’s infrastructure under Google’s enterprise privacy commitments, which for Workspace accounts include explicit commitments against training on customer data. For consumer Gmail accounts, the privacy picture is more complex, but the AI features operate within Google’s existing data framework rather than extending access to a new third party.
Third-party AI email clients are a different matter. They add a new party — the email app company — and potentially a second new party (their AI API provider) to the list of entities that process your email content. The privacy commitments of these additional parties may be stronger or weaker than Google’s, but they’re certainly different, and they’re often less scrutinized.
The practical question when evaluating an AI email client: “Which companies now have access to my email content, and what are their specific privacy commitments?” The answer is usually more than one.
Professional and Legal Implications
Two categories of users face particular risk from AI email clients: professionals who use personal email for work-related correspondence, and people who receive sensitive professional correspondence at their personal email.
For professionals: Attorneys, physicians, accountants, and financial advisors who use personal AI email clients with their work email may be routing client documents through AI processing services in ways that conflict with their professional obligations. Bar associations and medical boards have begun issuing guidance on AI tools, but compliance at the individual practitioner level varies widely.
For everyone: If you’ve ever received a legal document, medical record, or financial statement via email, it’s in your inbox. If you use an AI email client, it’s been processed by that client’s AI pipeline. Whether that creates legal risk depends on the specific documents, jurisdictions, and terms involved — but it’s a question worth being conscious of.
What to Do With Sensitive Email Documents
The better question isn’t “should I use an AI email client” — it’s “should I keep sensitive documents in my email inbox for years, making them accessible to every tool that reads my email?”
Email was not designed as a document archive. It became one because it was convenient — every document you received also has a timestamped copy in your inbox automatically. But an email inbox is an extremely permissive storage environment: it’s routinely accessible to email apps, backup services, productivity tools, and now AI clients.
Sensitive documents deserve dedicated storage with more deliberate access control. That means actually downloading and saving documents when you receive them, in a storage system designed for privacy rather than convenience:
- Bank statements belong in a private file storage system, not indefinitely in your inbox
- Medical records belong in a private health document store, not in a searchable email thread
- Legal documents belong in a private archive where you control access, not in an inbox that every new productivity app you try will promptly index
The documents in your email inbox are already there. The question is whether you want to keep adding new AI tools that read them, or whether you’d rather have a clear, private place for the documents that matter.
Where This Is Going
AI email clients are improving quickly, and the pressure to integrate AI features is now felt across the entire email ecosystem — not just specialized AI clients but mainstream email apps adding AI features as expected table stakes.
The privacy trajectory for email attachments depends on how users respond. Apps that compete on privacy — processing on-device, minimizing data transmission, explicit opt-in for AI features on sensitive document types — exist but are currently a minority. Most AI email development is focused on capability, with privacy as a secondary concern.
The specific question of whether attachment processing constitutes a meaningful privacy risk depends on what’s in your attachments. For users who receive routine email and don’t use their inbox as a document archive, the risk is modest. For users whose inboxes contain years of medical correspondence, financial records, and legal documents, the risk is more significant — and worth taking seriously before granting the next AI email app access to your entire inbox.
daftei is built to be the destination for important personal documents — a private, encrypted store separate from your email inbox and separate from any AI processing pipeline. Documents stored in daftei are encrypted with AES-256 at rest and transmitted with TLS 1.3. No AI reads them to build features or improve models. If you receive a document that matters, putting a copy somewhere outside your email inbox is worth the thirty seconds it takes.