To choose an email security solution, test it on your own mail against 12 checks, starting from the built-in Microsoft 365 baseline, which Microsoft documents with a spam effectiveness SLA above 99%². AegisAI is one example of the API model: it reads mail through the Microsoft 365 or Google Workspace API after delivery¹⁷. Each one ends with the question to put to a vendor.

What you can do by the end

Start from the built-in baseline

Microsoft says the built-in security features for all cloud mailboxes help protect a Microsoft 365 organization from email threats, and are on by default¹. Microsoft documents a spam effectiveness SLA above 99% and a false positive ratio SLA below 1:250,000 for those features.

AegisAI's inbound email security page prints one figure for that mail: 3 in 10 confirmed phishing emails cleared native Gmail and Microsoft filters in Q3 2026²³. Count that share on your own tenant before you buy.

Question to ask: how many confirmed threats did the product find in mail the built-in filter had already delivered, and how was each one confirmed?

Which deployment fits your stack: SEG or API-based email security?

An API-based tool fits a stack that must keep mail routing as it is. A secure email gateway (SEG) fits a stack that wants every message to pass through a filter before delivery. A gateway receives mail through an MX record change, so it sits in the delivery path. An API tool connects to Microsoft 365 or Google Workspace through an app the admin approves, and routing stays as it is.

The choice also decides what happens during an outage. If a gateway goes down, inbound mail waits for it. If an API tool goes down, mail keeps arriving while detection pauses. The guide on how to replace a secure email gateway without a cutover weekend covers running both side by side.

Question to ask: does deployment need a DNS MX change, and what happens to inbound mail if your service is down?

Which integration requirements apply in Microsoft 365 and Google Workspace?

Both platforms require an admin grant before a third-party tool can read mail. Microsoft calls admin consent on behalf of an organization a sensitive operation, because it can give the app's publisher access to significant portions of the organization's data¹³.

Revocation differs by platform. In Microsoft Entra ID, an admin revokes organization-wide permissions from the enterprise application's Admin consent tab with the Revoke permission option¹⁴. In Google Workspace, Revoke access in the Data Access section removes all access for an admin-installed Marketplace app, and access cannot be revoked for specific scopes¹⁵. Keep the read grant and the removal grant apart, so the tool can be watched before it can delete anything. The checklist on how to evaluate API-based email security goes deeper on permissions.

Question to ask: which permissions does the app request, which of them can move or delete mail, and how do we revoke each one?

How to judge the best AI email security platform: AI-native or traditional detection?

Judge an AI email security platform by the evidence it shows for each verdict.

Ask for three kinds of evidence. First, a reason for each verdict, in plain language, that an analyst can check against the message. Second, results on your own mail from a trial that ran in monitoring mode. Third, the denominator behind any accuracy figure: missed threats out of all threats, and false alarms out of all flagged mail. The page on how to evaluate detection accuracy and the false negative rate sets out that arithmetic. The explainer on how AI-powered email security works covers the models themselves.

Question to ask: for one blocked message and one allowed message from our trial, what reason did the model give, and which signals did it use?

Ask how one bad email is removed from every mailbox

Removal should reach every copy of a message, forwards included. It should also run on a grant the admin controls. When a tool finds one bad message, it needs to find and pull the other copies. Each removal belongs in a log. Removal is a write action, so it belongs on a separate grant that the admin approves after the monitoring period.

Question to ask: when you remove a message, do you remove the forwarded copies too, and where is each removal logged?

Which compliance and email security protocols belong on the checklist?

The three domain authentication records belong on it (SPF, DKIM and DMARC), because they let receivers check that mail claiming your domain really came from you. NIST SP 800-177 Rev. 1 names the three as mechanisms for authenticating a sending domain⁵. Microsoft says these DNS records let Microsoft 365 protect against spoofing more accurately⁴.

Google Workspace Admin Help says SPF helps prevent senders from impersonating you⁶. It also says DMARC lets you choose what happens to messages that fail SPF or DKIM. A security product does not replace these records. It should read their results as one signal per message, and keep the record for audits. The guide on how to roll out DMARC without breaking mail covers the records themselves.

Question to ask: does the product read the authentication results for each message, and does it flag mail that passes all three checks but still impersonates a supplier?

How to test 2026 threat readiness for BEC, zero-payload phishing and deepfakes

Test readiness with messages that carry no link and no attachment, such as a payment-change request. The FBI IC3 recorded 24,768 BEC complaints in 2025, against 21,442 in 2024⁸. Reported BEC losses for 2025 were $3,046,598,558. Businesses also reported losses over $30 million to BEC scams involving AI.

Deepfakes move the same request into voice and video. The FBI says criminals use AI-generated audio to impersonate public figures or personal relations to get payments¹¹. It also says criminals generate video for real-time chats with alleged company executives. A zero-payload email has no file to detonate and no URL to score, so detection has to read the request itself. Who is asking for the money, and does the request fit the sender's history? The guide on how to prevent business email compromise covers the payment controls around it.

Question to ask: how did the product handle a payment-change request with no link or attachment during our trial?

Test vendor figures on your own mail

Treat every catch rate in a brochure as a claim until your trial reproduces it. A figure without a test method and a denominator cannot be checked. Run the trial in monitoring mode first, so the product reports verdicts and moves nothing. Then compare its findings with the mail the built-in filter delivered over the same weeks. The page on how to structure a read-only email security proof of concept lays out the steps.

Question to ask: what method and denominator produced this figure, and can we reproduce it on our own mail?

What to look for in email security software, approach by approach?

Compare four approaches on the same five criteria: mail routing, timing, access, outages and undo steps.

How to score the best email security solutions with a 12-point checklist

Score each product yes or no on these 12 questions, using evidence from your own trial.

How AegisAI handles this

Through the Microsoft 365 or Google Workspace API, with an OAuth grant your admin controls and can revoke²⁴. There is no MX change and nothing in the delivery path. Runs in monitoring mode first¹⁹. Removal needs write access your admin grants and can revoke. Every verdict comes with its reason in plain language, so your team can check it and share it.

How to measure it in 30 days

Track three numbers from the first day of the trial.

Questions people ask

Sources