AegisAI Perspectives · 157
Stop phishing that Gmail misses: turn on the Admin console settings that ship off, enforce DMARC, then add a post-delivery API check with no MX change.
See it on your own Google Workspace or Microsoft 365 tenant. No MX change.
Or get the Q3 2026 State of the AI Attack report (free, instant download)
Not ready for a demo? Check your domain.

Gmail displays warnings and moves untrustworthy email to the spam folder by default1. To stop what it misses, turn on the Admin console protections that ship off, enforce SPF and DKIM, then DMARC. Then add a post-delivery check through the Google Workspace API. AegisAI connects to Microsoft 365 or Google Workspace through the API, judges every message by its intent, and removes what fails from every mailbox it reached73.
If you run IT for a small business or a company of 20 to 500 people on Google Workspace, with a small IT or security team and no SOC, this guide gives you the full admin path in one place. AegisAI reports that 3 in 10 confirmed phishing emails cleared native Gmail and Microsoft filters in Q3 2026, on its inbound email security page70. Four shorter pages answer single questions in more depth: the Gmail security settings checklist for admins, whether Google Workspace's built-in email security is enough, how a software startup stops the AI phishing Gmail misses, and what an API layer adds to Gmail's built-in protection.
Short on time? Watch the explainer, or book a demo.
| Question | Answer | Source |
|---|---|---|
| How do I stop phishing in Google Workspace that Gmail's built-in filters miss? | Turn on the protections that ship off, such as Security Sandbox, enforce SPF and DKIM, then DMARC33, and add a check that reads mail through the Google Workspace API after delivery, with no change to MX records6465. | Google Workspace Admin Help; inbound email security page |
| What does Gmail do by default? | It displays warnings and moves untrustworthy email to spam, and it scans all messages for malware. | Google Workspace Admin Help |
| Where are the phishing settings in the Admin console? | Apps, Google Workspace, Gmail, Safety, with the Gmail Settings administrator privilege2. | Google Workspace Admin Help |
| How do I set up SPF, DKIM and DMARC? | Publish an SPF TXT record at your domain host22, generate a DKIM key in the Admin console and start authentication2528, then publish a DMARC record that starts at p=none. | Google Workspace Admin Help |
| What still gets through? | Mail that passes authentication from a domain the attacker controls, because DMARC checks alignment with the sending domain. Business email compromise drew $3,046,598,558 in reported losses in 202559. | Google Workspace Admin Help; FBI IC3 |
| Does a post-delivery API layer need an MX change? | No. It reads mail through the Google Workspace API after delivery, and mail routing does not change. | Inbound email security page |
| How do I test one without risk? | Start read-only, in monitoring mode, and grant write access for removal only when you decide to666768. | Inbound email security page |
Gmail protects every Google Workspace mailbox from the first day, with no setting to change. It displays warnings and moves untrustworthy email to the spam folder. It scans all messages for malware, whether or not you turn on the attachment security settings. Google states that the AI defenses in Workspace automatically block more than 99.9% of spam, phishing attempts and malware from reaching users57. The defaults are a floor. The same Google guide points admins to enhanced pre-delivery message scanning, advanced phishing and malware protection and Security Sandbox as extra protections. Read the defaults below before you change anything, so you know what you are adding to.
It warns or moves the message to spam. Google's own words for the default are that Gmail displays warnings and moves untrustworthy emails to the spam folder. Nothing is deleted, and the user still sees spam if they open the folder. That matters for phishing. A user who searches spam for a missing invoice can still open the lure.
The warning banner has a limit of its own. Warning banners appear only in Gmail web, and third-party apps do not display them9. A user who reads mail in a desktop client or a phone app outside Gmail can see the same message with no banner at all.
Check which mail clients your users run. If many read Gmail outside the web client, treat the banner as absent for them and plan the settings below with that in mind.
Yes. Google scans all messages to protect against malware, whether or not the attachment security settings are turned on. That scan runs for every user on every edition, and you cannot switch it off.
The scan looks for malicious files only. A request to change bank details carries no file and no link, so a malware scan has nothing to find in it. That is why the rest of this guide covers Gmail's settings first, then a separate check of intent after delivery.
Two extra file checks sit on top of the base scan. Security Sandbox runs attachments in a virtual environment16. Enhanced pre-delivery message scanning holds suspicious mail briefly for more checks13. Leave the base scan as it is, and add the two checks where your edition allows.
They flag risky attachments and leave the email in the inbox with a warning. There are three attachment settings: protect against encrypted attachments from untrusted senders, protect against attachments with scripts from untrusted senders, and protect against anomalous attachment types in emails. The default action for each one is to keep the email in the inbox and show a warning.
A warning is a request to the user. The user can ignore it. An encrypted file cannot be opened by a scanner, which is why the first setting names it.
Decide the action per setting. For most small teams, moving the message to spam or to quarantine is safer than a warning for encrypted attachments and script attachments from untrusted senders. The settings section below shows where to change the action.
Two kinds. The first is the safety banner on a message Gmail judges risky. The second is the external recipient warning. External recipient warnings are on by default: they show a warning banner and an image or colored border next to external addresses49. You manage them at Apps, Google Workspace, Gmail, End User Access, Warn for external recipients.
The external warning helps on outbound replies. A user who replies to a lookalike domain sees that the address is outside the organisation. It does not help on inbound mail from a compromised partner account, because that partner is outside the organisation every day and the warning shows every day.
Leave external recipient warnings on. Do not count on any banner as the control that stops a wire fraud request, because the user decides what to do with it.
Yes. Automatic forwarding is on by default48. The setting is at Apps, Google Workspace, Gmail, End User Access, Automatic forwarding. The box is called Allow users to automatically forward incoming email.
An attacker who signs in to one mailbox can add a rule that copies every message to an outside address, and the user never notices. The setting is a door that stays open unless you close it.
Ask who needs automatic forwarding for work. If the answer is nobody or a short list, clear the box at the top level and turn it back on only for an organisational unit that needs it. Review the decision each quarter when roles change.
Several of the stronger ones. Gmail Security Sandbox is supported in Frontline Plus; Business Standard and Business Plus; and Enterprise Standard and Enterprise Plus15. The Enhanced malware and phishing protection setting, which turns on Enhanced Safe Browsing, is available in Business Standard and Business Plus; and Enterprise Standard and Enterprise Plus. The security investigation tool is supported in Frontline Plus; Enterprise Plus; and Education Standard and Education Plus52. Gmail log events in that tool are available only with Frontline Plus; Enterprise Plus; and Education Plus.
A small business on a starter edition gets none of these three. A business on Business Plus gets Sandbox and Enhanced Safe Browsing but not the investigation tool.
Write down your edition before you plan. Every setting in this guide carries its edition limit where Google states one, so you can mark what applies to you.
It covers spam, phishing attempts and malware as one total. The figure is a share of all three together, so it says little about the one targeted message that reaches a finance team.
The inbound email security page reports a different count: 3 in 10 confirmed phishing emails cleared native Gmail and Microsoft filters in Q3 202670. The two numbers measure different things. One is a share of all unwanted mail blocked. The other is a share of confirmed phishing that got through.
Plan for the targeted message that asks for money. The section on business email compromise below gives the FBI loss figures.
Turn on every Gmail safety setting your edition supports, set the action for each one, and close the doors attackers use after a sign-in. The table below lists each setting with its Admin console path, its default where Google states one, and its effect. Changing the Safety page settings requires the Gmail Settings administrator privilege. Apply settings at the top-level organisational unit first, then make exceptions for units that need them. For a shorter checklist of the same settings, see the Gmail security settings checklist.
| Setting | Admin console path | Default | Effect | Source |
|---|---|---|---|---|
| Attachment protections (3 settings) | Apps > Google Workspace > Gmail > Safety | Keep in inbox, show warning | Flags encrypted attachments, script attachments and anomalous types from untrusted senders | |
| Links and external images (3 settings) | Apps > Google Workspace > Gmail > Safety | Check in your console | Identifies links behind short URLs, scans linked images, warns on clicks to untrusted domains | 5 |
| Spoofing and authentication (5 settings) | Apps > Google Workspace > Gmail > Safety | Check in your console | Acts on lookalike domains, employee name spoofing, spoofing of your own domain, unauthenticated mail | 6 |
| Enhanced pre-delivery message scanning | Apps > Google Workspace > Gmail > Spam, Phishing and Malware | Check in your console | Slightly delays suspicious mail for additional security checks | 14 |
| Security Sandbox (Enable virtual execution) | Apps > Google Workspace > Gmail > Spam, Phishing and Malware | Off | Analyzes Office documents, .exe files and PDFs in a sandbox | |
| Enhanced malware and phishing protection | Apps > Google Workspace > Gmail > Spam, Phishing and Malware | Off | Turns on Gmail Enhanced Safe Browsing | 51 |
| Automatic forwarding | Apps > Google Workspace > Gmail > End User Access | On | Lets users forward incoming mail automatically | |
| Warn for external recipients | Apps > Google Workspace > Gmail > End User Access | On | Banner and border next to external addresses | |
| Third-party app access | Security > Access and data control > API controls | Check in your console | Sets each app to Trusted, Limited or Blocked | 45 |
Two pages hold them. The Safety page protections (attachments, links, spoofing) are at Menu, Apps, Google Workspace, Gmail, Safety, and changing them requires the Gmail Settings administrator privilege. Enhanced pre-delivery message scanning is at Menu, Apps, Google Workspace, Gmail, Spam, Phishing and Malware, where you select your top-level organization and check the box. Security Sandbox and Enhanced Safe Browsing sit on that same page.
Give the Gmail Settings administrator privilege to the one or two people who own mail settings, not to every IT helper. A wrong action on a spoofing setting can send real supplier mail to quarantine, so changes should go through someone who knows what each setting does.
Open both pages now and screenshot the current state of every setting. That screenshot is your baseline for the 30-day checklist at the end of this guide.
Turn on all three and choose a stronger action than a warning for the two that name untrusted senders. Each of the three defaults to keeping the email in the inbox with a warning.
The action you pick can be Quarantine. When you select Quarantine for any advanced security setting, the quarantine applies only to incoming messages, and allowlist settings do not override it10. That second point is useful. A sender on an allowlist still gets the quarantine, so a compromised partner on your list does not slip past.
Start with Move to spam for anomalous attachment types, where false positives are more likely, and Quarantine for encrypted and script attachments from untrusted senders. Watch the quarantine for two weeks and adjust.
Turn on all three. The links and external images settings identify links behind shortened URLs, scan linked images, and show a warning prompt for any click on links to untrusted domains.
Each one closes a common trick. A shortened URL hides the real destination from a quick look, and the first setting resolves it. A linked image can carry the lure text as a picture so that text scanning finds nothing, and the second setting scans it. The click warning gives the user one more prompt before a page they have never visited loads.
Each click warning is a prompt the user can dismiss. Pair these settings with the spoofing settings below, because a lure that reaches the inbox from a trusted-looking sender gets more clicks. Turn the three link settings on at the top level today; they rarely cause false positives.
Turn on all five. The five settings protect against domain spoofing based on similar domain names, spoofing of employee names, inbound emails spoofing your domain and any unauthenticated emails, and protect Groups from inbound emails spoofing your domain.
Read what each covers. Protect against inbound emails spoofing your domain covers messages not authenticated with either SPF or DKIM that pretend to be from your domain7. The setting protect against any unauthenticated emails requires messages to be authenticated by any domain, with SPF or DKIM or both, and shows a question mark next to the sender name when they are not8.
Neither setting judges mail that does authenticate. A lookalike domain the attacker registered can pass SPF and DKIM for itself. The similar domain names setting is the one aimed at that case. Set the employee name and similar domain settings to move mail to spam or quarantine, and the unauthenticated setting to a warning first.
Yes, for most organisations. With enhanced pre-delivery message scanning, when Gmail detects suspicious content, message delivery is slightly delayed so that Gmail can do additional security checks on the message. To turn it on, go to Menu, Apps, Google Workspace, Gmail, Spam, Phishing and Malware, select your top-level organization, and check the Enhanced pre-delivery message scanning box.
The cost is a short delay on suspicious mail only. Normal mail is not held. For a finance team, a short delay on a message Gmail already finds suspicious is a fair trade.
Tell your help desk before you switch it on. A user who expects a password reset email and sees it arrive late will call them. Give the help desk one line to say: suspicious mail gets extra checks, and it arrives shortly. Then turn it on at the top level.
Yes, if your edition has it. Gmail Security Sandbox is supported in Frontline Plus; Business Standard and Business Plus; and Enterprise Standard and Enterprise Plus. It analyzes Microsoft Office documents, Microsoft executables (.exe) and PDFs; scanning adds a slight delay, and the scan should be complete within a few minutes.
The setting is off by default. You turn it on at Menu, Apps, Google Workspace, Gmail, Spam, Phishing and Malware by checking the Enable virtual execution box.
A sandbox opens the file in a virtual environment and watches what it does. That catches a document whose code only runs when opened, which a static scan can miss. It does nothing for a message with no file. Turn it on for every user, then check after a week whether any team reports slow delivery of attachments they need fast, and scope it if so.
Yes, if your edition has it. The Enhanced malware and phishing protection setting turns on Gmail Enhanced Safe Browsing; it is off by default, and it is available in Business Standard and Business Plus; and Enterprise Standard and Enterprise Plus. To turn it on, go to Menu, Apps, Google Workspace, Gmail, Spam, Phishing and Malware, scroll to Enhanced malware and phishing protection and check the box, which requires the Gmail Settings administrator privilege.
Sandbox and Enhanced Safe Browsing both carry an edition limit, and the list of Business and Enterprise editions is the same for each.
Turn it on at the top level. Then note it in your baseline screenshot, so a later audit can see the date it changed.
Yes, in most cases. Each advanced security area has an option to automatically turn on and apply future recommended settings12.
The benefit is that new protections arrive without you reading every release note. The risk is that a new setting changes how mail is handled on a day you did not plan for. For a team with no SOC, the benefit is larger. You will not have time to track each new Gmail protection by hand.
Turn the option on for attachments, for links and external images, and for spoofing and authentication. Add a monthly task to open the Safety page and note what changed, so that a sudden rise in quarantined mail has an explanation you can find.
Mostly no, and that is by design. The advanced security features generally work independently of other spam settings, so even a domain listed as a safe sender in spam settings still has the enhanced security features applied11. When an advanced setting is set to Quarantine, allowlist settings do not override the Quarantine option.
Approved senders who bypass spam classification are a separate case. Google recommends requiring authentication for approved senders who bypass spam classification, because when sender authentication is off Gmail cannot verify the message was sent by the person it seems to come from55.
Open your spam settings and find every approved sender list. For each list, check that Require sender authentication is on. An approved sender without authentication is a gap an attacker can use simply by forging that sender's address.
For most users, yes. Google recommends disabling POP and IMAP access for users who do not explicitly need it, because IMAP clients might not have protections similar to first-party clients56. Since May 1, 2025, Google Workspace accounts no longer support less secure apps, third-party apps or devices that sign in with a username and password, and these apps must use OAuth47.
The two points work together. OAuth removed password sign-in for old clients. Turning off IMAP for people who do not need it removes the remaining clients outside Gmail's own protections.
List who uses an IMAP client today. Turn POP and IMAP off at the top level, then turn it back on in a small organisational unit for the people with a real need, and review that unit each quarter.
Publish SPF at your domain host. Then generate and start DKIM in the Admin console. Last, publish DMARC at p=none and move it to enforcement as reports allow. SPF needs nothing in the Admin console19. DKIM needs a super administrator in the Admin console. DMARC needs SPF or DKIM or both working first31. The table lists every step in order, with where you do it and how long to wait.
| Step | Where | What to do | Wait before the next step | Source |
|---|---|---|---|---|
| 1. List your senders | Your own inventory | Name every service that sends mail as your domain: web servers, on-premise servers, outbound gateways, automated mail, third-party providers | None | 20 |
| 2. Publish SPF | Domain host | Add a TXT record, host @, value such as v=spf1 include:_spf.google.com ~all | Up to 48 hours to take effect | 1823 |
| 3. Wait for the DKIM key | Admin console | After Gmail is turned on, wait before the key is available | 24 to 72 hours | 24 |
| 4. Generate the DKIM key | Apps > Google Workspace > Gmail > Authenticate email | As super administrator, select the domain, click Generate New Record, pick 2048 bit if your host supports it | None | 26 |
| 5. Publish the DKIM key | Domain host | Add a TXT record, host google._domainkey, value v=DKIM1; k=rsa; p=key | Until the record is live | 27 |
| 6. Start DKIM | Apps > Google Workspace > Gmail > Authenticate email | Select the domain, click Start authentication; status reads Authenticating email with DKIM | None | |
| 7. Verify DKIM | Another Gmail or Workspace account | Send a message, open Show original, look for DKIM=pass | None | 29 |
| 8. Set up a report mailbox | Admin console groups or a mailbox | Create a group or mailbox for DMARC reports | None | 32 |
| 9. Publish DMARC at none | Domain host | Add a TXT record, host _dmarc.yourdomain, with p=none | At least 48 hours after SPF and DKIM authenticate | 36 |
| 10. Enforce | Domain host | Change p to quarantine, then to reject, over time | As reports allow | 34 |
Each answers a different question about a message that claims to be from your domain. SPF lists the servers allowed to send for your domain; the ~all tag tells receiving servers to mark mail from unlisted servers as spam21. DKIM signs each message with a key whose public half you publish in DNS as a TXT record. DMARC tells receiving servers what to do with messages from your domain that fail SPF or DKIM, and the options are reject, quarantine or deliver30.
DMARC adds one more test: alignment. A message passes DMARC if it passes SPF with SPF alignment, or DKIM with DKIM alignment. Alignment means the domain that passed is the domain in the From address the user sees.
All three protect your own domain from being forged. They do not judge mail from other domains. Set up all three for every domain you send from, including domains used only for marketing or billing.
No. Google states that you do not need to do anything in your Google Admin console to set up SPF: you determine your SPF record, then sign in to your domain host and add it.
The work is in the inventory. Step 1 of SPF setup is to identify all your email senders, such as web servers, on-premise mail servers, outbound gateways, services that send automatic email and any third-party provider that sends email for your domain. A small business often finds senders it forgot: a billing tool, a website contact form, a help desk, a marketing platform.
Make the list before you touch DNS. Ask finance, sales, support and whoever runs the website what tools send email with your domain in the From line. Every sender you miss will start to fail SPF once DMARC is enforced, and its mail will go to spam or be rejected.
Start from Google's example and add your other senders. The example SPF record for a domain that sends email with Google Workspace only is v=spf1 include:_spf.google.com ~all. Each other sender on your list adds its own include or address before the final tag.
The final tag sets the policy for unlisted servers. The ~all tag tells receiving servers to mark messages as spam if they come from servers not listed in the SPF record, and Google recommends ~all.
To publish it, create or update a TXT record at your domain host, with the host set to @ for the same domain and the value set to the SPF record. Keep one SPF record per domain; merge any existing record into the new one. It can take up to 48 hours for SPF authentication to start working. Note the time you published it, because DMARC depends on that clock.
Not on day one of a new tenant. In Google Workspace, after you turn on Gmail for your organization, you must wait 24 to 72 hours before you can get your DKIM key in the Admin console.
For an existing tenant that has run Gmail for months, the key is available now. For a new tenant, put the DKIM step on your calendar for three days after Gmail goes live, and use the wait to finish your sender inventory and SPF record.
Check whether DKIM is already on. Open Apps, Google Workspace, Gmail, Authenticate email, and select your domain. If the status already reads Authenticating email with DKIM, DKIM is running and you can move to the verify step. If it is not, generate the key as the next section describes.
Generate it in the Admin console, then publish it at your domain host. You must be signed in as a super administrator to generate a DKIM key, and the path is Menu, Apps, Google Workspace, Gmail, Authenticate email; select the domain and click Generate New Record.
Pick the key length. DKIM key bit length options are 2048 and 1024, and Google says to select 2048 if your domain provider supports it, because longer keys are more secure.
Then publish it. Step 2 of DKIM setup is to add a TXT record at your domain host with the host name google._domainkey (the prefix selector followed by ._domainkey) and a value formatted v=DKIM1; k=rsa; p=key. Copy the value from the Admin console in one piece, with no added spaces or line breaks.
Turn it on in the Admin console, then test from another account. Step 3 of DKIM setup is to go back to Authenticate email and select the domain, then click Start authentication, after which the status changes to Authenticating email with DKIM.
To verify DKIM, send a message to someone using Gmail or Google Workspace, open the message header with Show original, and look for DKIM=pass in Authentication-Results. You cannot verify DKIM by sending yourself a test message. A test to a personal Gmail account you own works well.
If the header shows a fail, check first for a TXT value that was cut or changed when pasted. Compare the published record with the value in the Admin console character by character. Repeat DKIM setup for every domain and alias domain in your tenant.
You need working SPF or DKIM and a place for reports. You must turn on SPF and/or DKIM before you can use DMARC, and Google says to allow 48 hours after setting up SPF and/or DKIM before setting up DMARC.
Reports come next. Step 1 of DMARC setup is to set up a group or dedicated mailbox to receive reports, because large organizations might get up to hundreds or even thousands of reports daily. A small domain receives fewer, and each one is an XML file that is hard to read by eye.
Create a group such as dmarc-reports at your domain. Decide who reads it, and how. A free report parser or a spreadsheet import is enough for a small tenant. The reports are the evidence you need later to move from none to enforcement, so do not skip this step.
Add one TXT record at your domain host. Step 4 of DMARC setup is to add a TXT record with the host _dmarc.example.com (your domain) and the DMARC record as the value. DKIM and SPF should be authenticating messages for at least 48 hours before you turn on DMARC.
Start with p=none. When you start using DMARC, Google recommends setting the policy option p to none, then over time changing the receiver policy to quarantine or reject. With p=none, the receiving server delivers the message and logs it in a daily report. Nothing is blocked, so nothing breaks.
Include a reporting address that points at the report group you created, so the daily reports arrive there. Publish the record, and set a reminder to read the first two weeks of reports before you change anything.
When the reports show that every legitimate sender passes. The p tag has three values: none delivers the message and logs it, quarantine sends it to the recipient spam folder, and reject rejects the message. Google's advice is to start at none and change to quarantine or reject over time.
Read the reports for senders that fail. Each one is either a service you forgot to add to SPF or sign with DKIM, or someone forging your domain. Fix the first kind. The second kind is what enforcement is for.
Move to quarantine first, then watch for a week or two for complaints about missing mail. Then move to reject. A domain at reject tells every receiver to refuse forged mail in your name, which protects your customers and suppliers as well as your own staff. The DMARC steps are rows 8 to 10 of the table above.
They apply to anyone who sends to Gmail accounts. Since February 1, 2024, all senders to Gmail accounts must set up SPF or DKIM and have valid forward and reverse DNS records37. They must also use a TLS connection and keep spam rates reported in Postmaster Tools below 0.30%37.
Bulk senders have more to do. Senders of more than 5,000 messages per day to Gmail accounts must set up all three records (SPF, DKIM, DMARC), and the DMARC enforcement policy can be set to none38.
If your domain sends fewer than 5,000 messages a day, the first set of rules is the one that applies. A marketing platform sending newsletters as your domain can push you over the line on its own. Count your daily volume across every sender on your list. If you are near the limit, set up all three records now rather than when mail starts to bounce.
It does not stop mail from domains you do not own. DMARC protects your own domain. A message passes DMARC if it passes SPF or DKIM with alignment to its own From domain. An attacker who registers a lookalike domain can publish all three records for it, and the message passes them.
It also does not stop mail from a real account that was taken over. That mail comes from the right servers and carries a valid signature, so authentication passes.
Gmail's own setting for unauthenticated mail has the same limit. Protect against inbound emails spoofing your domain covers messages not authenticated with either SPF or DKIM. A message that authenticates is outside its scope. Enforce DMARC, and plan a separate control for mail that authenticates and still lies.
Mail that authenticates, carries no malicious file, and asks for something a real colleague or supplier might ask for. Gmail's defaults and settings judge files and links, plus sender authentication. A message that passes all three can still be a fraud.
Because authentication checks the sender's own domain, not whether that domain is the one the user expects. A message passes DMARC when SPF or DKIM passes with alignment to the From domain. If the From domain is a lookalike the attacker owns, alignment holds.
Gmail has a setting aimed at this case: protect against domain spoofing based on similar domain names. It is a useful layer. It works on similarity, so a lookalike that is not close enough to any domain your users know, or a free mailbox with a convincing display name, can still pass.
The user sees a familiar name and a domain that looks right on a phone screen. Turn on the similar domain names setting with a spam or quarantine action. Train finance staff to read the full address on any payment request. Then add a check that reads what the message asks for, which the later sections describe.
Because there is nothing in it for a file or link scan to find. Google scans all messages for malware, and the link settings act on links and linked images. A message that says "please update our bank details for the next invoice" carries neither.
The loss from this kind of mail is large. Business email compromise accounted for 24,768 complaints and $3,046,598,558 in reported losses to the FBI in 202559. Total reported losses that year reached $20.877 billion58.
Inside Gmail, the employee name spoofing setting helps when the attacker uses a known name from outside the organisation. It does not help when the request comes from a real supplier account. Write down a call-back rule for every change of bank details. Then add a check that reads the intent of the message.
Because it is genuine mail from a genuine account. When an attacker signs in to a supplier's mailbox, every message they send comes from the supplier's real servers with the supplier's real signature. All three authentication checks pass. The setting for unauthenticated mail does not apply, because the mail is authenticated.
The message often arrives inside an existing thread. The attacker replies to a real invoice conversation and changes one detail. The external recipient warning shows the address is outside your organisation, which is true of every message from that supplier.
Nothing in the Gmail settings judges whether the request is normal for that supplier. That judgement needs the history of the thread and the sender. Add a payment change rule that requires a call to a number already on file, and plan an after-delivery check that compares a request with past mail from the same sender.
Because a banner asks the user to decide, and some users never see it. The default action for each attachment protection is to keep the email in the inbox and show a warning. Warning banners appear only in Gmail web, and third-party apps do not display a warning banner.
A user in a hurry reads past a banner. On a phone mail app outside Gmail, the banner never shows. Staff who see the same external warning on every supplier email learn to ignore it.
Use banners as information, not as the control. For the settings where a false positive is rare, such as script attachments from untrusted senders, change the action from a warning to spam or quarantine. Keep the banner for cases where you want the user to look twice and the risk of blocking real mail is higher.
That phishing is the most reported crime and business email compromise is among the most costly. IC3 received 1,008,597 complaints in 2025, with $20.877 billion in reported losses, a 26% increase in losses from 202458. Phishing and spoofing had 191,561 complaints in 2025, the highest count of any listed crime type, with $215,843,126 in reported losses60.
The 2025 report also counts complaints tied to AI. Complaints with an AI Related descriptor numbered 22,364 in 2025, with $893,346,472 in reported losses61.
Read the numbers together. Phishing is the most frequent complaint, but each report costs little on average. Business email compromise is rarer and costs far more per complaint59. For a small team, the message to plan for is the one that asks for money, not the one that carries malware.
Because the tool that finds and removes a message from every inbox is not in every edition. The security investigation tool is supported in Frontline Plus; Enterprise Plus; and Education Standard and Education Plus, and Gmail log events are available only with Frontline Plus; Enterprise Plus; and Education Plus.
A business on Business Starter, Standard or Plus does not have the investigation tool. When one user reports a phishing email, the admin cannot search every mailbox for the same message from one screen. Each user has to find and delete it, or the admin uses whatever search their edition allows.
Check your edition against this list now. If you lack the investigation tool, write down how you would remove one bad message from every mailbox today, and how long it would take. That time is the gap a post-delivery layer is meant to close.
With the investigation tool, if your edition has it. In the tool you can identify all users in your domain that received a malicious message, delete it from their Gmail inboxes, mark it as spam or phishing, or send it to the user inbox53. The tool is at Menu, Security, Security center, Investigation tool; it requires the Security center administrator privilege, and log data might take up to a few minutes to become available54.
Practise once before you need it. Pick a harmless test message sent to a few colleagues, search for it by subject and sender, and run the delete action on a test group. Note how long the whole task took.
If your edition does not include the tool, the answer depends on a person. Write a short runbook that names who searches, who tells users and who confirms removal. Then compare that time with what an after-delivery API layer does automatically, covered in the next section.
It connects to Google Workspace through the API, reads each message after Gmail delivers it, and removes the ones that fail. Mail routing stays as it is. The inbound email security page states it plainly: the layer reads mail through the Microsoft 365 or Google Workspace API after delivery, and there is no change to mail routing (MX records). Gmail keeps doing its job first. The API layer is a second reader that sees what Gmail let through. For the short version of what this adds to Gmail, see what an API layer adds to Gmail's built-in protection.
| Scope or control | What it allows | Where you check it | Source |
|---|---|---|---|
| gmail.readonly | View email messages and settings; a restricted scope | The app's entry in API controls | |
| gmail.modify | Read, compose and send email; no immediate permanent deletion that bypasses trash | The app's entry in API controls | 40 |
| Restricted scope rules | OAuth App Verification, plus a security assessment if the app stores the data on servers | Google's verification of the app | 41 |
| Domain-wide delegation | Access to all users' data without each user's consent | Security > Access and data control > API controls > Manage Domain Wide Delegation | 4244 |
| App access: Trusted | All Google Workspace services, including restricted ones | Security > Access and data control > API controls | |
| App access: Limited | Unrestricted services only | Security > Access and data control > API controls | |
| App access: Blocked | No Google service | Security > Access and data control > API controls |
Where the layer sits. An MX record tells the internet which server receives mail for your domain. Changing it sends every inbound message through another service before Gmail. A post-delivery API layer leaves MX records alone and reads mail through the Google Workspace API after delivery.
The difference shows up on the day you set it up and on the day something fails. With no MX change, there is no DNS cutover and no risk of mail bouncing during setup. Gmail's own filters still run first, so nothing you configured in the earlier sections is bypassed.
The trade-off is timing. A layer after delivery acts on a message that is already in the inbox, so its speed matters. Ask any provider for the time from delivery to verdict and from verdict to removal, and check both during a trial.
Through OAuth, with a grant the admin gives and can take back. The AegisAI home page describes an OAuth grant your admin controls and can revoke75. Third-party apps now reach Workspace accounts through OAuth. Since May 1, 2025, Google Workspace accounts no longer support apps that sign in with a username and password, and you must use OAuth.
A tool that reads every mailbox can use domain-wide delegation. Domain-wide delegation lets you grant client applications permission to access your Workspace users' data without requiring their consent. A super administrator sets it up at Menu, Security, Access and data control, API controls, Manage Domain Wide Delegation, by clicking Add new, entering the Client ID and adding each OAuth scope.
Before you add any client, read its scope list line by line.
Usually one Gmail scope to read, and one to act. The gmail.readonly scope is described as view your email messages and settings, and it is listed as a restricted scope. By contrast, gmail.modify is described as read, compose, and send emails from your Gmail account, and it does not allow immediate, permanent deletion of threads and messages, bypassing the trash.
A layer that only reports what it finds needs read access. Removal from inboxes needs a scope that can change messages. Ask any provider which scopes it requests in each mode, and check that list against what you see in the Admin console when you add the client.
Watch for scopes outside Gmail. Each scope on the list should have a stated reason. If a scope has no reason you can write down in one line, ask before you grant it.
Read-only can look; modify can change. With gmail.readonly, an app can view email messages and settings. Under gmail.modify, it can read, compose or send email.
The modify scope matters for removal. A layer with gmail.modify that removes a phishing email leaves it recoverable from trash. That gives an admin a way back if a real message is removed by mistake.
Both scopes are sensitive. Google lists restricted Gmail scopes as providing wide access to Google user data, and requires restricted scope OAuth App Verification; an app that stores restricted scope data on servers must go through a security assessment. Start any trial with gmail.readonly, and add gmail.modify only when you decide to allow removal.
Access to every user's data for the scopes you list. With domain-wide delegation, the app has access to the data belonging to all of your users, and Google recommends a regular review of service accounts and deleting any no longer in use43.
That is a large grant. The control is in the scope list and in the review. A client with gmail.readonly can read mail for every user; it cannot send or change any message, because that scope only lets it view messages and settings.
Put a review on the calendar. Once a quarter, open Manage Domain Wide Delegation, list every client and its scopes, and delete any you no longer use. Note the date and who did it. For more on what a mailbox API grant means in practice, see email security API access and mailbox permissions.
By setting each app to Trusted, Limited or Blocked. Third-party app access is managed in the Admin console at Menu, Security, Access and data control, API controls, which requires the Service Settings administrator privilege. A trusted app can access all Google Workspace services including restricted ones; Limited can access only unrestricted services; Blocked cannot access any Google service.
This is the page to visit before and after any trial. Before, it shows which apps already have access to your users' mail. After, it is where you block an app you no longer use.
Open API controls today and sort the list by access level. Any app marked Trusted that you do not recognise is the first item to check. Ask its owner what it does, and block it if no one can say.
Mail keeps flowing. Because the layer is not in the mail path, an outage on its side does not stop delivery. The inbound email security page states it in two short sentences: mail keeps flowing, and detection pauses until the service returns69.
That is the main operational difference from a layer that sits in front of Gmail. With an MX change, every inbound message depends on the front layer being up. An API layer outage leaves Gmail delivering as normal; only the second reader stops.
Gmail's own protections keep running during an outage, because they never depended on the API layer. Ask any provider how it catches up on mail delivered while it was down, and how you will know an outage happened. Write both answers into your runbook.
Fast enough that the speed has to be measured, not assumed. A layer after delivery acts on a message already in the inbox, so the gap between delivery and removal is the window in which a user can open it. The inbound email security page states that the median verdict takes 1.9 seconds, and a convicted message is removed from every mailbox it reached within seconds71.
Compare that with the manual path. Without the investigation tool, which is limited to some editions, one admin searches and deletes by hand. With it, log data can take a few minutes to become available before the search shows every copy.
During a trial, measure the time from delivery to verdict and from verdict to removal from the last mailbox. Do it on your own mail, with your own clock.
Grant read access only, let the layer report what it would have done, and compare its findings with what Gmail caught. The inbound email security page describes this start: the layer starts read-only and runs in monitoring mode first. Removal needs write access your admin grants and can revoke. A read-only test changes nothing in any mailbox, so it is safe to run on real mail. To run one on your own tenant, ask for a monitoring-mode session.
The layer can view mail and cannot change it. With gmail.readonly, an app can view email messages and settings. It cannot move, label or delete a message, because those actions need a scope that can modify mail.
In monitoring mode, the layer reads each delivered message and records a verdict and a reason72. Nothing happens to the user's inbox. The output is a list of messages the layer would have removed, with the reason for each.
That list is the evidence a small team needs. It shows, on your own mail, what Gmail let through and why the second reader flagged it. Agree with the provider before you start how you will receive the list, and how often.
Add the client with the read scope only, then check it in two places. A super administrator adds the client at Menu, Security, Access and data control, API controls, Manage Domain Wide Delegation, entering the Client ID and each OAuth scope. For a read-only trial, the Gmail scope on that list should be gmail.readonly.
Then check API controls. Third-party app access is at Menu, Security, Access and data control, API controls. Find the app and note whether its access setting is Trusted, Limited or Blocked. A Limited app can access only unrestricted services. If your organisation has restricted Gmail in API controls, ask the provider which access setting the trial needs.
Screenshot both pages on the day you grant access. Those screenshots are your record of exactly what the trial could do.
Minutes, on the provider's own figures. The platform page states: no MX record change, and about five minutes to connect74. The demo request page states it is an API connection to Microsoft 365 or Google Workspace that takes 5 minutes, and mail routing does not change76.
On the admin side, the scope review and the screenshots in the previous section take longer than the grant itself.
Plan for both. Book an hour with the person who holds the super administrator role. Use the first part to read the scope list and agree what the trial will measure, and the rest to grant access and check it.
Four numbers. First, how many messages the layer flags that Gmail delivered to the inbox. Second, how many of those your team confirms as phishing or fraud after reading the reason. Third, how many flagged messages were real mail, which is the false positive count. Fourth, the time from delivery to verdict.
Use the reasons to confirm verdicts. A person can read the reason and decide quickly whether the verdict is right. Keep a simple sheet with four columns for the message, the verdict, the reason and your call. Two weeks of that sheet is enough to decide.
Line the two lists up by message. For each message the layer flags, check whether Gmail warned, moved it to spam or left it untouched. A message Gmail left untouched with no banner is the clearest case of something it missed.
If your edition includes the security investigation tool, use it to search Gmail's view of the same message: who received it and what Gmail did with it. Log data might take up to a few minutes to become available.
Look at the patterns, not only the count. Are the missed messages lookalike domains, supplier accounts or requests with no link? Each pattern points back to a setting or a gap in the earlier sections of this guide. Fix the settings you can fix in Gmail first, so the trial result reflects the gap that remains.
When the read-only list shows verdicts your team agrees with, and the false positive count is one you can live with. In Gmail terms, write access is a scope that can change messages, such as gmail.modify.
Grant write access in stages. Start with one organisational unit, such as finance, where the cost of a missed fraud is highest. Watch the removals for a week and check the reasons. Then extend it to everyone. Keep the read-only screenshots and add new ones for the day you added the write scope.
In two places, in a few minutes. Remove the client from Manage Domain Wide Delegation, and set the app to Blocked in API controls, so it cannot access any Google service. Google recommends deleting service accounts that are no longer in use.
The connection was built to be revocable. The home page describes an OAuth grant your admin controls and can revoke.
Write the revoke steps into your trial plan before you start, with the name of the person who will do them. If you decide not to continue, revoke on the last day of the trial and screenshot both pages to show the access is gone.
Settings in week 1, authentication in weeks 1 to 3, access review and a read-only test in weeks 2 to 4. The order follows the dependencies in this guide: SPF needs up to 48 hours to work, DKIM can need 24 to 72 hours on a new tenant, and DMARC waits 48 hours after both.
Day 1: screenshot every setting on both Gmail safety pages, and write down your edition.
Days 1 to 3: turn on every protection on both pages that your edition supports.
By day 3: list every service that sends mail as your domain, then publish SPF.
Days 3 to 5: generate and publish the DKIM key. Then start authentication and check for DKIM=pass.
Days 5 to 7: set up a DMARC report group, then publish DMARC at p=none once SPF and DKIM have run for 48 hours.
Week 2: review automatic forwarding; POP and IMAP; approved senders; and API controls. Block apps you cannot explain.
Weeks 2 to 4: run a post-delivery API layer read-only, with gmail.readonly only, and record each verdict and reason.
Day 30: decide on DMARC enforcement and on write access for removal, using the report data and the trial sheet.
Settings and SPF. Start with the baseline screenshots, because every later decision compares against them. Then turn on the protections that ship off: Security Sandbox is off by default, and so is Enhanced Safe Browsing. Turn on enhanced pre-delivery message scanning at the top level, and set the actions on the attachment and spoofing settings.
Run the sender inventory at the same time. Step 1 of SPF setup is to identify all your email senders. Publish SPF by the middle of the week, because it can take up to 48 hours to start working.
If your tenant is new, check the DKIM key is available. It can take 24 to 72 hours after Gmail is turned on. By the end of week 1, SPF and DKIM should both be authenticating, and the DMARC record should be live at p=none.
Close the doors an attacker uses after a sign-in, and start the access review. Automatic forwarding is on by default; turn it off for everyone who does not need it. Google recommends disabling POP and IMAP access for users who do not explicitly need it. Check that every approved sender list requires sender authentication.
Then open API controls. List every app and whether it is Trusted, Limited or Blocked. Open Manage Domain Wide Delegation and list every client and its scopes. Delete any you no longer use, as Google recommends.
Read the first week of DMARC reports. Each sender that fails is either a service missing from SPF or DKIM, or a forgery. Fix the missing services this week, so that the data in week 4 is clean.
Run the read-only test and prepare the two decisions. Grant a post-delivery API layer read access only. With gmail.readonly, an app can view messages and settings and cannot change them. Let it run in monitoring mode and keep the trial sheet.
One person can review the list each morning. Compare flagged messages with what Gmail did with them, and use the investigation tool where your edition has it.
By day 30 you have two sets of data. The DMARC reports tell you whether every legitimate sender passes, which decides whether you move from none to quarantine. The trial sheet tells you what reached inboxes after Gmail, which decides whether you grant write access for removal.
AI agents investigate every inbound email for what it is trying to do, and remove the ones that fail from every mailbox they reached63. Removal needs write access your admin grants and can revoke68. Every verdict comes with its reason in plain language, so your team can check it and share it72.
One IT manager can finish this plan in 30 days, with settings and SPF in week 1, the access review in week 2 and a read-only test in weeks 3 and 4.
Illustrative example, figures are fictional: Harbor Lane Dental Group (fictional) has 140 people on Google Workspace Business Plus. It has one IT manager and no SOC. In week 1 the IT manager finds Security Sandbox and Enhanced Safe Browsing both off, which matches the documented defaults, and turns them on. The sender inventory finds a billing tool and a booking system that send as the practice's domain. Both go into SPF, and DMARC goes live at p=none. In week 2, API controls show two unused apps marked Trusted; both are blocked. In weeks 3 and 4, a post-delivery layer runs read-only. It flags 9 delivered messages that Gmail left in inboxes. The IT manager confirms 6 as phishing from the reasons, including a supplier request to change bank details that passed all three authentication checks. Three are newsletters, logged as false positives. On day 30, DMARC moves to quarantine and write access is granted for the finance unit first.
Track six numbers, each from a page you can open.
Put the six side by side on day 30. They give you the evidence for the DMARC enforcement decision and the write access decision in the checklist.
Turn on the protections that ship off, such as Security Sandbox, set the spoofing settings, and enforce DMARC. Then add a check after delivery. AegisAI reads mail through the Google Workspace API after delivery, with no change to MX records, and starts read-only.
Yes. Google scans all messages to protect against malware, whether or not attachment security settings are turned on. The scan looks for malicious files. A request with no file or link, such as a bank detail change, gives it nothing to find.
No. The Security Sandbox setting is off by default; you turn it on at Apps, Google Workspace, Gmail, Spam, Phishing and Malware by checking Enable virtual execution. It is supported in Frontline Plus; Business Standard and Business Plus; and Enterprise Standard and Enterprise Plus.
No. Google states you do not need to do anything in the Admin console to set up SPF: you determine the record, then add it at your domain host. DKIM is different and needs a super administrator in the Admin console.
You must wait 24 to 72 hours after you turn on Gmail for your organization before you can get your DKIM key in the Admin console. On an existing tenant, the key is available now.
Start with p=none. Google recommends p=none at the start, then changing the policy to quarantine or reject over time. With none, the message is delivered and logged in a daily report, so nothing breaks while you read the reports.
No. A post-delivery layer reads mail through the Google Workspace API after delivery, and there is no change to mail routing (MX records). Gmail's own filters still run first.
The gmail.readonly scope lets an app view email messages and settings. With gmail.modify, an app can read, compose or send email, and it does not allow immediate, permanent deletion that bypasses the trash. Use readonly for a trial.
Mail keeps flowing. Detection pauses until the service returns. The layer is not in the mail path, so Gmail delivers as normal and its own protections keep running.
Delete its client in Manage Domain Wide Delegation, and set the app to Blocked in API controls, so it cannot access any Google service. Google recommends deleting service accounts that are no longer in use.