Opens in a new tab
Back

Is email secure enough for sending sensitive data, and what does email DLP actually add?

Email is encrypted hop by hop, and misdirected mail tops UK breach reports. What HIPAA and PCI DSS allow, and what email DLP catches and misses.

Table of Contents

A waiter in a white jacket offers a wax-sealed envelope to a woman in a jazz club who raises her palm to refuse it, while a man at another table raises his hand

For most sensitive data, email is acceptable with safeguards, and PCI DSS forbids sending card numbers unprotected. Mail between large providers is almost always encrypted in transit, but that encryption is hop-by-hop and usually opportunistic. It does nothing when the message reaches the wrong person. Email sent to the wrong recipient was the largest category among the UK ICO’s 13,457 breach reports for 2025. Corporate email DLP can inspect an outgoing message and warn, block or quarantine it. A spreadsheet attached in a personal Gmail tab never reaches that corporate mail flow, so closing that gap takes an endpoint agent that sees uploads and pastes on the laptop itself. Jazz does this at the operating-system level, in any browser and without an extension, and Melody investigates whether the movement fits the person’s work. Encryption and mail routing still belong to the mail system.

Updated October 23, 2026.

Which tools can stop this, and what each one sees

Misdirected work mail and work data leaving through personal webmail are different paths, and each control sees only part of them. Purview’s email DLP runs inside Exchange Online and comes with Microsoft 365 E3. Proofpoint has a gateway and API-based Adaptive Email DLP. Both depend on the corporate mail systems connected to them, and a personal Gmail tab can send the same spreadsheet without passing through either. Jazz records that browser upload or paste on the laptop, while the mail system handles encrypted delivery.

Situation Control Covers Misses
Work mail in Exchange Online Mail-flow inspection; Purview is included with E3 Outbound content policies and supported encryption actions Personal-webmail uploads and pastes
Work mail in Google Workspace Gmail DLP on a supported edition Warn, block and quarantine inside Gmail Unsupported editions and personal mailboxes
Several mail systems Inspection across the mail routes Messages that pass through those routes Personal mail and any internal mail outside the route
Misdirected email Recipient checks, a send delay and anomaly detection Addressing mistakes caught before delivery Familiar wrong addresses and mail already delivered externally
Work mail forwarded to personal accounts Automatic external forwarding turned off Inbox rules that copy mail outside Messages a person sends by hand, which need a content-and-recipient DLP rule
Required transit encryption Enforced TLS or message encryption Transmission on the routes in scope A wrong recipient with permission to decrypt
Personal webmail blocked at the network A proxy that decrypts TLS, as Google’s consumer-account block requires Google sign-ins on devices that use the proxy and trust its certificate Outlook.com, Yahoo and devices off the tunnel
Personal webmail controlled in the browser A browser extension or managed browser Uploads and pastes in the browsers it manages Other browsers, portable browsers and desktop apps
Work data attached or pasted in personal webmail Jazz’s endpoint agent on the work laptop Uploads and pastes in any browser with no extension, investigated by Melody, with a response that fits the risk Devices without the agent and Linux machines; desktop-mail coverage needs a separate check

Is email encrypted in transit and at rest?

Email between major providers is encrypted in transit almost all the time now. Google’s Transparency Report showed about 99% of the mail Gmail sent to other providers, and nearly all the mail it received from them, encrypted with TLS on October 2, 2026. A USENIX Security 2023 study found that more than 96% of reachable mail servers for the top 10 million domains accept TLS. Pages that still say email travels in plain text by default are out of date.

That encryption covers less than the padlock suggests. STARTTLS, defined in RFC 3207 in 2002, encrypts one hop between two servers, and the RFC itself says “SMTP is not an end-to-end mechanism.” NIST SP 800-177 Rev. 1, published in February 2019, describes mail moving “hop-by-hop over a series of relays” and warns that a bad or privileged actor “can read such mail on the submission or delivery systems.” The ICO’s guidance on encryption in transit adds that without further encryption the data “will be stored in the same plaintext form on the recipient’s system.”

It is also opportunistic. Microsoft says Exchange Online “always attempts to send messages using TLS,” but when the destination doesn’t support it, the default is to send the email anyway, and Gmail does the same. An attacker who deletes the “250 STARTTLS” line from the conversation, the downgrade RFC 3207 itself describes, gets the same result. Senders rarely check certificates either. In a 2025 study presented at ACM IMC, 93.2% of the 2,394 sending domains in its sample accepted any certificate and only 1.3% always required a valid one. The 2023 USENIX Security study found 30% of presented certificates invalid and all 47 providers it tested connecting to them anyway.

Two standards let a receiving domain demand validated TLS: DANE (RFC 7672, 2015) and MTA-STS (RFC 8461, 2018). NIST’s Security Recommendation 5-3 says receivers “should implement DANE, MTA-STS (or both),” and CISA’s Binding Operational Directive 18-01 gave federal agencies 90 days in 2017 to offer STARTTLS on every internet-facing mail server. Adoption elsewhere is thin. The IMC study found MTA-STS records on 1.2% of the 10,000 most popular domains it examined, 29.6% of roughly 68,000 MTA-STS domains misconfigured, and only 19.6% of its sending domains checking MTA-STS when delivering to a domain that publishes it.

You can still require encryption on the routes that matter. Exchange Online’s Outbound messages in Transit Security report shows the recipient domains that got your mail in plain text because they didn’t support TLS. Gmail’s secure transport (TLS) compliance setting can require TLS, a CA-signed certificate and a matching hostname for named domains, and since March 2026 Exchange Online’s outbound connectors offer a mandatory DANE mode that queues mail failing validation. If PHI or card data goes to the same partner every week, require TLS on that route instead of relying on the default.

At rest, the big providers encrypt stored mail by default. Microsoft says Microsoft 365 data “is always encrypted at rest” with BitLocker, with service encryption as a second layer. Google says Gmail encrypts customer data at rest and in transit by default. Unless you add client-side encryption, the service can use the keys, which is how it can search and scan your mail. That protects disks and data centers. A message sitting in the wrong person’s mailbox opens for them like any other.

End-to-end options keep the content encrypted from sender to recipient:

  • S/MIME (RFC 8551) and OpenPGP (RFC 9580, July 2024) encrypt the message to the recipient’s key. NIST’s Recommendation 5-5 prefers S/MIME with a CA-signed certificate for federal use.
  • Portal, or “pull,” encryption sends a link, and the recipient reads the message on the sender’s encrypted web portal.
  • Rights-managed messages, such as Microsoft Purview Message Encryption or Google Workspace client-side encryption, tie access to an identity. Google says client-side encrypted mail uses keys “not available to Google servers.”
  • Gmail’s confidential mode replaces the content with a link, so the sender can revoke access after a mistake, but Google warns that recipients “can still take screenshots or photos.”

A password-protected attachment sits between these. The ICO’s encryption guidance says to send the password by a separate channel, such as a phone call once delivery is confirmed, and “Do not include the password within the same email as the encrypted file attachment.” Done that way, it still protects a file sent to the wrong address, because the wrong recipient never gets the password. It is also a file DLP can’t read.

End-to-end encryption blinds DLP in the same way. Anything encrypted on the sender’s device reaches Exchange, Gmail or a gateway as ciphertext, and a scanner without the key can read the headers and little else. Google’s own FAQ says DLP scans “can’t access client-side encrypted content in files or email,” and Microsoft’s Exchange DLP reference describes password-protected attachments as content that “can’t be scanned.” The workable order is inspect first, then encrypt. Purview can apply Message Encryption as the action of a DLP rule after the content has been checked, and Proofpoint’s email DLP can encrypt outbound mail as a policy action. Gmail’s DLP rules have no encrypt action.

Where email leaks actually happen

Regulators publish better numbers here than any vendor survey, and those numbers show people sending the right data to the wrong place.

The ICO publishes the personal data breach reports it receives as a downloadable incident-trends file. Counting distinct reports for calendar 2025, “data emailed to incorrect recipient” was the largest single incident type, with 2,459 of 13,457 reports (18.3%), well ahead of phishing at 1,526. “Failure to use bcc” added another 434, so about one report in five was an email addressing mistake. Another 788 reports from 2025 had no incident type recorded, so those counts are a floor. Education and childcare filed the most misdirected-email reports, 415, followed by health with 405 and legal with 346, and in legal, misdirected email made up 36% of everything the sector reported. The ICO’s own annual report for 2025/26, published in July 2026, says sending personal information to the wrong person by email, post or fax “continued to be” the most common incident type, at “just over 20% of breaches.”

US healthcare shows the same failures. HHS’s annual breach report to Congress for 2024 names email as the main location of the PHI in 164 of the 663 breaches of 500 or more people that occurred that year (25%), second only to network servers. Among smaller breaches, it lists emails “sent to the wrong individuals” as one kind of misdirected communication. On HHS’s breach portal, which lists only breaches affecting 500 or more people, 195 of the 802 reports filed in 2025 named email as a location. Most of those (158) were hacked mailboxes, which phishing defenses and MFA address. Several of the 37 filed as unauthorized disclosures describe staff emailing PHI where it shouldn’t go. Fidelis Care New York reported a workforce member who “emailed documents containing the protected health information (PHI) of 1,680 individuals to the wrong recipient.” Aprendamos Family of Services reported a workforce member who put 1,996 patients’ records in an Excel spreadsheet and “emailed it to her personal email account prior to her departure.”

The Verizon 2026 Data Breach Investigations Report, published in May 2026, says “Misdelivery is also the top error type in the overall dataset (61%),” and asks whether an organization has “a way of intercepting these kinds of errors before there is a full-blown data breach.” In its DLP data, personal webmail was the target of 20% of 4,280,149 untrusted DLP events by non-malicious insiders, second only to file-sharing sites at 32%. Describing breaches driven by convenience, it gives the example of an employee who “wants to work from home and emails company data to a personal account.”

The fines follow the same pattern. The ICO fined the Ministry of Defence £350,000 in December 2023 after its Afghan relocation team put 245 people’s addresses in the To field of an email instead of BCC. Three such emails exposed 265 addresses, and the ICO’s starting figure was £1 million. Central YMCA was fined £7,500 in 2024, cut from a recommended £300,000, after a coordinator used CC on an email to 270 addresses on an HIV support program. The next day’s Outlook recall attempt “led to another email to all 270 recipients.” In 2016 the NHS trust that runs London’s 56 Dean Street clinic was fined £180,000 after a newsletter for users of its HIV service exposed 781 recipients’ email addresses, according to the ICO’s press release. In July 2026 the ICO issued the Metropolitan Police a reprimand and enforcement notice after an officer emailed 18 crime victims with their addresses in the To field. The ICO also said BCC itself “was not a suitably secure method” there, because special category data about the victims could be inferred.

The failures differ, and each needs its own control:

  • Autocomplete picks a similar name, often someone already in the sender’s contacts.
  • Reply-all exposes the whole thread to everyone on it.
  • To or CC goes where BCC was meant, or a group email goes out that should have been sent one by one.
  • The right person gets the wrong attachment.
  • An auto-forwarding rule copies work mail to a personal account.
  • Someone sends a spreadsheet to their own personal address to finish it at home.

Is email secure for sensitive data under HIPAA, PCI DSS, GLBA and GDPR?

Most rules allow email with safeguards chosen by risk assessment and written down. PCI DSS is the strict one. The FTC’s guide for business, dated October 2016, says “Regular email is not a secure method for sending sensitive data” and recommends encrypting any such transmission. The table quotes the operative text of each rule.

Data and rule What the text says Can you email it?
PHI under the HIPAA Security Rule The Security Rule “does not expressly prohibit the use of email for sending e-PHI” (HHS FAQ). Encryption in transit is an addressable specification under 45 CFR 164.312(e), so you assess it, choose and document the decision. HHS’s FAQ 570 suggests “checking the e-mail address for accuracy before sending.” Yes, with documented safeguards. A patient who asks for PHI by unencrypted email can have it after “a brief warning” about the risk (HHS). HHS’s January 2025 proposal to require encryption isn’t final, and the federal regulatory agenda lists final action for July 2027.
Card numbers (PAN) under PCI DSS v4.0.1 Requirement 4.2.2 Requirement 4.2.2 “prohibits the sending of unprotected primary account numbers (PANs) via end-user messaging technologies,” email included (PCI SSC FAQ 1085, August 2025). Mail systems that carry PAN come into PCI scope (FAQ 1310). Only with strong cryptography, and the mail servers involved join your PCI scope.
Customer financial information under the FTC Safeguards Rule (GLBA) Encrypt customer information “in transit over external networks and at rest.” Where that’s infeasible, use compensating controls “reviewed and approved by your Qualified Individual” (16 CFR 314.4(c)(3)). Yes, encrypted, or with a documented and approved alternative.
Nonpublic information under NYDFS 23 NYCRR 500.15 A written policy requiring encryption “both in transit over external networks and at rest.” Since the November 2023 amendment, the compensating-controls exception covers data at rest only (NYDFS). Yes, encrypted.
Personal data under GDPR and UK GDPR Article 32 Security appropriate to the risk, “including inter alia as appropriate” the “pseudonymisation and encryption of personal data” (UK GDPR Article 32). For sensitive personal information, the ICO says “you should use alternatives to BCC” (ICO). Yes, scaled to the risk. Bulk and special-category mail needs more than BCC.
Massachusetts residents’ personal information under 201 CMR 17.04(3) Encrypt “all transmitted records and files containing personal information that will travel across public networks,” to the extent technically feasible (201 CMR 17.00). The state’s FAQ says that if it isn’t feasible, you needn’t, but best practice is “not sending unencrypted personal information in an email.” Encrypt where feasible.
Client confidences under ABA Formal Opinion 477R (2017) “The use of unencrypted routine email generally remains an acceptable method of lawyer-client communication,” with special precautions when the information is highly sensitive (ABA). Yes for routine matters. Encrypt or use another channel for highly sensitive ones.
Student education records under FERPA The Education Department’s Student Privacy Policy Office says “FERPA does not have an explicit information security standard as it is a privacy, not an information security, law,” and names encrypted attachments and SFTP as safer ways to send records (US Department of Education, March 2020). Yes, where FERPA permits the disclosure. Encrypting attachments is the recommended practice.

These rules are about interception. Transport encryption and most message encryption decrypt for whoever the message is addressed to, so a wrong recipient reads it like the right one would have. An encrypted attachment whose password goes by phone is the exception, and it’s a file DLP can’t inspect. In the BCC cases the ICO has fined, such as the Ministry of Defence’s £350,000 penalty, encryption would not have prevented the disclosure, because the data went to the addresses the sender typed. In practice, interception and misdirection need separate controls.

What email DLP does and where it runs

Email DLP reads outgoing mail, decides whether it holds something that shouldn’t leave, and acts before or as the message goes. The built-in tools inspect message bodies and attachments with pattern-based detectors (Social Security numbers, card numbers, custom regular expressions, keyword lists), trainable classifiers and sensitivity labels, and Gmail can read text in images with optical character recognition. Conditions combine content with context, such as “Recipient domain is” or “Attachment is password protected” in Microsoft Purview. The ICO’s audit toolkit lists the plainest version among the measures to consider: “Automatically quarantine outgoing emails containing sensitive information” (ICO).

The actions are similar across products:

  • Warn. Outlook shows a policy tip, and Gmail’s “Warn users” lets the sender continue and logs the choice. The EU’s Article 29 Working Party said in Opinion 2/2017 that a DLP tool should warn the sender before transmission, “so as to give the sender the option to cancel this transmission.”
  • Block, optionally with an override and a written justification.
  • Quarantine the message for an administrator to release or reject in Gmail, or deliver it to Microsoft 365’s hosted quarantine with Purview.
  • Ask a person. Purview can forward the message for approval to the sender’s manager or to named approvers.
  • Encrypt with Purview Message Encryption, optionally forcing outside recipients to read it in the encrypted portal.
  • Audit only, add a header or footer, or redirect the message.

What each one sees depends on where it runs.

Built-in DLP runs inside the mail service. Purview DLP for Exchange Online comes with Microsoft 365 E3, and our DLP software comparison breaks down what E3 and E5 each add. Google documents Gmail DLP for its Frontline, Enterprise and Education editions, and Business editions aren’t on that list. Gmail checks messages as they’re sent from Gmail on the web and the mobile app. Mail sent through a third-party client over SMTP is scanned after it leaves, and there “Warn users” behaves like a block.

A secure email gateway, such as Proofpoint, Mimecast or Cisco Secure Email, sits in the mail route and applies one policy to every mailbox and mail server routed through it. Proofpoint pairs its email DLP with encryption, and Mimecast lists secure messaging among its DLP features. A gateway scans what crosses it, and Cisco’s guide notes that “Messages sent to other users within your network are not scanned.”

API and add-in tools work from the mailbox. Proofpoint’s Adaptive Email DLP deploys through Microsoft 365 APIs, and Proofpoint, which agreed in October 2023 to buy Tessian for its misdirected-email technology, says the product learns from more than 12 months of a user’s mail and scores each outbound message on its recipients, relationships and attachments. A tool that scans only after delivery can report an external send but can’t take it back.

What email DLP misses, according to the vendors’ own documentation

Microsoft’s and Google’s documentation lists most of these gaps. The last three follow from how email works.

  1. Password-protected files. Microsoft’s Exchange DLP reference describes such attachments as ones that “can’t be scanned,” and detects the password on Office files, .zip, .7z and PDFs. Google lists “Contents of password-protected files” as not scanned. Both let you block or quarantine them as a class, which stops the legitimate ones too.
  2. Mail encrypted before it left the sender. Google says DLP scans “can’t access client-side encrypted content in files or email,” and an S/MIME or PGP message reaches any scanner as ciphertext unless the scanner holds the key.
  3. Links in place of attachments. Google’s size-limits page says “Content in the destination of URL links isn’t scanned,” and Drive files linked to a message fall under Drive’s DLP rules instead. Microsoft doesn’t say either way. A link into someone’s personal Dropbox or Drive sits outside your tenant altogether.
  4. Large files and messages. Gmail DLP scans messages and attachments up to 25 MB and analyzes only the first 10 MB of extracted text. Purview scans the first 2 million characters of a file and marks anything longer as not completely scanned.
  5. Mail already stored. Microsoft says DLP “does not scan or match previously existing email items that are stored in a mailbox or archive.”
  6. Rule logic and client gaps. Google warns that rules built on a NOT condition can give “unexpected results” when a message carries several attachments, because a rule fires if any single attachment matches. In Gmail, mail from third-party apps is scanned after it leaves and groups get only the Block action. Purview’s oversharing pop-ups need an E5-class license and don’t appear in Outlook on the web.
  7. Personal webmail, and every other route that skips your mail flow, from a personal Gmail tab on a work laptop to the mail app on someone’s own phone.
  8. Recipients your gateway never sees. A wrong internal recipient doesn’t cross an outbound gateway, and a rule that allows mail to an outside domain allows the wrong person at that domain too.
  9. Context. Pattern matching finds a Social Security number and stops there, with no way to tell your payroll provider from the sender’s personal inbox.

How to stop misdirected email, and what each control catches

Control What it catches What it misses Setting
Outlook External Recipients MailTip Any outside recipient, or a group that includes one, before sending A wrong outside address, which gets the same tip as the right one; a wrong internal recipient; the wrong attachment Off by default; turn on with Set-OrganizationConfig -MailTipsExternalRecipientsTipsEnabled $true
Outlook Large Audience MailTip A distribution group bigger than the threshold A long list of individual addresses; Microsoft documents the tip for distribution groups On by default, with a threshold of 25 members
Gmail external recipient warnings Outside recipients in a thread, a reply or a new message Outside addresses in the Directory or Contacts, and anyone the user has emailed before, whom Gmail saves to Other contacts. Autocomplete suggests those same addresses, so a familiar wrong one gets no warning On by default; CISA’s Gmail baseline GWS.GMAIL.13.1v1 says they “SHALL be enabled”
Undo send The sender’s second thoughts during the delay Anything noticed later Gmail 5 to 30 seconds; Outlook for Mac up to 120 seconds; new Outlook and Outlook on the web a short delay; classic Outlook rules that defer delivery up to 120 minutes
Outlook recall Mail inside your Microsoft 365 organization, or in tenants whose admins allow-listed yours Gmail and every other outside system; copies the recipient already forwarded by hand or by inbox rule Cross-tenant recall is off by default
DLP policy tips and oversharing pop-ups Sensitive content heading outside A plausible but wrong recipient for content that may leave Oversharing pop-ups need E5-class licensing and don’t run in Outlook on the web
Recipient-anomaly tools A recipient or attachment that doesn’t fit the sender’s history A new relationship with no history; Mimecast’s tool skips recipients the sender has already emailed twice Learn from past mail
Gmail confidential mode Lets the sender remove access after a mistake Screenshots and photos; the subject line, which is still sent Chosen per message before sending; replaces the body and attachments with a link

The ICO’s email security guidance adds process. The ICO published it in August 2023 and says it is under review after changes made by the Data (Use and Access) Act. For group mail that is sensitive, it points to “bulk email services or mail merge services” instead of BCC, and it suggests warning senders who use the CC field, setting a delay before mail leaves, and turning off the autocomplete function. HHS’s FAQ 570 suggests confirming a patient’s address before sending PHI to it. The Metropolitan Police’s numbers show how often such a prompt fires once it’s on. The force told the ICO that since March 11, 2026 it has shown staff a prompt whenever an email has more than 10 external recipients, and 2,969 users had seen it 5,995 times.

Recall deserves the least trust. Microsoft’s recall documentation says “You can’t recall messages over the internet,” the ICO’s guidance describes an NHS trust staff member who “attempted, unsuccessfully, to recall the messages,” and Central YMCA’s recall sent a second email to all 270 recipients.

People keep emailing spreadsheets to personal Gmail. What can block that?

Two different paths look the same in an incident report, and only one of them passes through email DLP.

Work mail sent to a personal address

When a spreadsheet goes from the work mailbox to an @gmail.com address, the mail flow sees it and email DLP can act. Three controls do most of the work.

Turn off automatic external forwarding. Microsoft changed the default of its outbound spam policy to “Off” in 2021 for new organizations and for those not using the older setting, but tenants already relying on it can still forward, so Microsoft advises setting the value explicitly. Google Workspace leaves user auto-forwarding “on by default” until an admin turns it off. CISA’s Exchange Online baseline says automatic forwarding to external domains “SHALL be disabled” (MS.EXO.1.1v2), and its Gmail baseline says it “SHOULD be disabled” (GWS.GMAIL.11.1v1). Our guide to spotting a departing employee taking data lists the audit events that show forwarding rules people have already created.

Match content and recipient together. A DLP rule that warns or blocks when customer data goes to consumer domains catches the spreadsheet. A blanket block on @gmail.com also stops freelancers, job applicants, patients and small partners whose only address is a consumer one, and each of them becomes another exception to maintain. Blocking personal email on its own also pushes the same files toward file-sharing sites and AI tools, as our guide to stopping employees leaking data explains.

Watch for the sender’s own address. HHS’s 2025 filings include PACE Greater New Orleans, whose employee “emailed the protected health information (PHI) of 715 individuals to her personal email account,” and Mayo Clinic, where a workforce member sent 1,869 people’s PHI to a personal address and disclosed it further. Proofpoint says its Adaptive Email DLP detects attempts to send sensitive data to personal accounts.

Personal webmail opened in a browser

When someone opens personal Gmail in Chrome and attaches a work spreadsheet there, nothing touches the corporate mail flow, and no email DLP, built-in or gateway, sees it. In the Verizon 2026 Data Breach Investigations Report, personal webmail was the target of 20% of untrusted DLP events by non-malicious insiders. Two kinds of control can see this path.

An endpoint agent on the laptop sees the upload or paste before the browser encrypts it. Jazz’s endpoint agent records it at the operating-system level and works across every browser with no extension to install. Microsoft Purview Endpoint DLP, which needs E5 or an add-on, can restrict uploads to listed cloud domains and sends people using an unallowed browser to Microsoft Edge.

A proxy that decrypts TLS sees the upload when the device is on its tunnel and trusts its certificate. Outside ChromeOS, Google’s block on consumer accounts works this way, with a proxy that must “support SSL interception” and add an X-GoogApps-Allowed-Domains header. Plain URL filtering can’t do the job, Google explains, because “legitimate traffic from a user’s managed Google Account goes to the same URL as the traffic you want to block.” The block covers Google sign-ins only, so Outlook.com and Yahoo need controls of their own.

The personal-webmail row of our endpoint, network and cloud DLP coverage matrix shows which layer sees what. Seeing that a file went to personal webmail doesn’t require reading anyone’s personal mailbox, and our guide to monitoring law explains why you should avoid it.

Why pattern rules misfire on email, and what judging by context looks like

A pattern rule checks one thing, whether a message contains something shaped like a Social Security number or a card number. HR’s monthly file to the payroll provider and a spreadsheet headed for a personal Gmail account give the same answer. Microsoft’s own Purview planning guidance describes internal auditors who “must share some credit card numbers with third-party auditors,” and says a policy that prohibits all external sharing would cause “a significant business process disruption.” Its remedy is an exception for named individuals or audit-only mode, and its policy design guide shows a block rule that adds “do NOT block the email if the recipient is in the contoso.com domain OR the fabrikam.com domain.”

The people who run these tools end up writing the business into exceptions. Pitt Digital excludes its hospital partner’s @upmc.edu addresses from email DLP notifications, and only from email (Pitt Digital). The Texas A&M University System reserves exceptions for roles “such as Financial Aid, the Registrar, or Human Resources” that have “a documented, legitimate business need” to transmit SSNs. Each exception is a recipient-domain or group rule that someone has to write and later remember to remove. The Article 29 Working Party also warned that a “false positive” alert “might result in unauthorized access of legitimate e-mails,” which makes false positives a privacy cost as well as a time cost.

Judging by context means asking what a reviewer would ask: what the data is, where it’s coming from and going to, who is sending it and whether that’s normal for them, and whether it fits a known business process. A payroll file that goes from the HR manager to the payroll provider’s domain on the first of every month is the business running as usual. If the same file goes from her account to a personal Gmail address late at night, two weeks after she resigned, nothing in the business explains it, and that is the one to stop. Our guide to writing DLP policies as flows shows how to write the auditor and payroll cases without a blanket pattern rule.

Test your email controls with fake data

Plant fake data, such as a spreadsheet of invented names with test Social Security numbers and card numbers that pass the Luhn check. Run each check from a managed laptop and record what each control saw and did.

  1. Email the file from your work account to a personal Gmail address, from Outlook desktop, Outlook on the web and a phone.
  2. Send it again inside a password-protected .zip.
  3. Upload it to OneDrive and email only the sharing link.
  4. Create an inbox rule that forwards mail to an outside address, and check whether it fires.
  5. Open personal Gmail in a browser, attach the file and paste a few rows into the body.
  6. Send a test message to 30 outside test addresses using CC, and confirm the warnings you expect appear.
  7. Try to recall a message sent to an outside test address.
  8. Address a third message to an outside test address you’ve already emailed twice, and note which warnings still appear.

The testing checklist in our DLP software comparison adds per-OS and SIEM checks.

How Jazz handles personal-webmail uploads

An attachment dragged into a personal Gmail tab is a browser upload, and rows pasted into the compose window are a paste. Jazz’s endpoint agent records both at the operating-system level, along with screenshots and GenAI prompts, and tells a personal account from a corporate one on the same service, as it describes for GenAI tools.

Melody judges each movement on four dimensions: what the data is, which systems it moved between, who moved it, and whether it fits a known business process (how Melody investigates). You don’t have to list approved recipient domains in advance. When several teams start sharing material data with a partner such as Deloitte, Melody asks whether that’s an acceptable business activity and remembers your answer (example). What reaches you is a pre-investigated incident that tells the whole story, and the response fits the risk, from a nudge or a request for justification to a targeted block. In a typical 30-day deployment, Jazz processes about 2 million signals and surfaces about 80 investigated incidents, according to Jazz.

Jazz doesn’t encrypt mail, sit in the mail route, scan mailboxes on the server or recall messages. A company that must encrypt PHI or card data in transit still needs required TLS or message encryption, and recipient warnings in the mail client, such as Outlook’s External Recipients MailTip and Gmail’s external recipient warnings, still matter. Coverage depends on the agent, so devices you don’t manage are out of scope. For desktop mail clients, ask Jazz for the per-OS coverage list, as our DLP software comparison recommends for every vendor.

Frequently asked questions

Is Gmail or Outlook email encrypted?

In transit, almost always between large providers. Google's Transparency Report showed about 99% of Gmail's outbound mail sent over TLS in early October 2026. That TLS is hop-by-hop and opportunistic, so each mail server holds a readable copy, and Exchange Online and Gmail both still deliver when the receiving server doesn't support TLS.

Is it a HIPAA violation to email PHI?

Not in itself. HHS says the Security Rule "does not expressly prohibit the use of email for sending e-PHI," but encryption is an addressable specification you must assess and document. Patients who ask for unencrypted email can have it after a brief warning. HHS's January 2025 proposal to require encryption isn't final; the federal agenda lists July 2027.

Can you send credit card numbers by email under PCI DSS?

Not unprotected. PCI SSC FAQ 1085 says Requirement 4.2.2 prohibits sending unprotected card numbers through end-user messaging, email included. If card numbers do travel by email, they need strong cryptography, and under FAQ 1310 the mail servers that carry them come into your PCI scope.

How common is misdirected email?

It is the most common breach type the UK ICO sees. In its 2025 data, email sent to the wrong recipient accounted for 2,459 of 13,457 breach reports, and failure to use BCC added 434. The Verizon 2026 Data Breach Investigations Report calls misdelivery "the top error type in the overall dataset (61%)."

Can you recall a misdirected email?

Only inside Microsoft 365. Outlook recall works within your organization, or in tenants whose admins have allow-listed yours, and Microsoft says you can't recall messages over the internet. Gmail has a 5- to 30-second undo window, and confidential mode lets you remove access only if you chose it before sending. Central YMCA's recall attempt emailed all 270 recipients again.

Does email DLP work on personal email opened in a browser?

No. A file attached in a personal Gmail tab never enters the corporate mail system. Jazz's endpoint agent records the upload or paste on Windows and Mac laptops across browsers, without an extension. Melody investigates the incident. A laptop without the agent is outside that coverage.

What's the best email DLP for stopping employees sending sensitive attachments to the wrong people?

It depends on the path the attachment takes. From the work mailbox, recipient warnings and a send delay catch some addressing mistakes before delivery, and a DLP rule that matches content and recipient together catches sensitive data heading to consumer domains. When the attachment goes out through personal Gmail opened in a browser on a work laptop, no email DLP sees it, and an endpoint agent is the control that does. Jazz records that upload or paste at the operating-system level in any browser, without an extension, and its response fits the risk: a nudge, a request for justification or a targeted block. Melody investigates whether the movement fits the person's work. Jazz doesn't sit in the mail route, so a misaddressed work email still needs the mail-side controls.

How do we stop staff attaching work files to personal webmail without a proxy or browser extension?

Put the control on the laptop. Outside ChromeOS, Google's block on consumer accounts needs a proxy that intercepts TLS, and it covers Google sign-ins only, so Outlook.com and Yahoo need controls of their own. Jazz's agent works at the operating-system level, so it sees the upload or paste in any browser with nothing installed in the browser, on the Windows and Mac laptops where it runs. Unmanaged devices and Linux machines need separate controls.

Should we use Microsoft Purview or a gateway such as Proofpoint for email DLP?

Purview inspects mail inside Exchange, while a gateway inspects messages routed through it. The choice depends on your mail systems and encryption requirements. A personal Gmail upload bypasses both. Jazz covers that upload or paste on the work laptop, alongside the controls used for corporate mail and encrypted delivery.

Does encrypting an email make it compliant?

It answers interception, which is what the encryption language in HIPAA, the Safeguards Rule, NYDFS and the Massachusetts rule addresses. It doesn't help when the message goes to the wrong person, who decrypts it like the right recipient would. An encrypted attachment whose password goes by a separate channel, as the ICO advises, is the exception.

Will blocking @gmail.com stop staff emailing data to personal accounts?

Partly, and at a cost. A blanket block also stops freelancers, job applicants, patients and small partners who only have consumer addresses, and it does nothing about personal webmail opened in a browser. On the mail-flow path, turn off automatic external forwarding and warn or block when sensitive content goes to consumer domains.

Does Jazz encrypt email or replace an email gateway?

No. Jazz doesn't encrypt mail, sit in the mail route, scan server-side mailboxes or recall messages, so PHI or card data still needs required TLS or message encryption. Its endpoint agent covers what email DLP misses, such as attachments uploaded to personal webmail in a browser, on the devices where it's installed.

Keep exploring