AegisAI Perspectives · 117
A modern email security platform reads intent, follows links, catches payment fraud and removes mail after delivery, and each one needs a test in a trial.
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 (free, about 10 seconds).
A modern email security platform needs eight capabilities, led by intent analysis and payment fraud detection, and each one needs a test on your own mail. BEC alone drew 24,768 FBI complaints in 20251. AegisAI's agents investigate every inbound email for what it is trying to do21. Run the tests below in a trial before you sign.
The platform reads what each message asks the reader to do, such as paying, signing in or sharing data. The FBI's IC3 notes that chat generators can quickly create official-sounding emails mimicking a company's CEO or other officials7.
The platform opens links and attachments before a user does, and checks links again when they are clicked. Microsoft's Safe Links adds time-of-click verification of URLs on top of scanning during mail flow18, because a link legitimate on delivery can be weaponized later.
The platform flags a request to move money even when the message carries no link or file. IC3 recorded $3,046,598,558 in BEC losses in 20252.
The platform reads DMARC results for each sender domain. DMARC tells receiving servers what to do with messages that fail SPF or DKIM authentication5.
The platform pulls a message out of every mailbox once it turns bad. Microsoft notes users still receive malicious messages, such as content weaponized after delivery17, and its zero-hour auto purge searches only the last 48 hours of delivered email.
Each verdict carries a reason an analyst can check, and the platform feeds existing SecOps tools such as SIEM and SOAR15.
Users send suspect mail to a reporting mailbox, which holds messages users mark as malicious or not malicious in Outlook20.
The platform keeps good mail flowing, because false positives increase as more products are added. No number of layers brings total protection to 100 percent.
Copy these eight lines into your trial plan as the rows you will score.
Short on time? Watch the explainer, or book a demo.
Payment fraud detection answers the BEC line in the FBI's IC3 report. IC3 counted 24,768 BEC complaints in 2025, up from 21,442 in 2024. Reported BEC losses reached $3,046,598,558, against $2,770,151,146 in 2024. In 2025, businesses reported losses over $30 million to BEC scams involving AI3.
A filter that scans only links and attachments has nothing to scan in a plain text request to change bank details. Phishing and spoofing complaints were a separate line, at 191,561 in 20254.
Seed the trial with a plain text bank change request from a personal address, sent to a finance mailbox you control.
A modern email security platform keeps acting after delivery, because some messages turn malicious later. Microsoft's own example is a link that was legitimate on delivery and later weaponized, which Safe Links did not detect. Its zero-hour auto purge searches only the last 48 hours of delivered email.
Platforms that connect through the Microsoft Graph API scan messages after delivery and remove the ones they judge malicious13. That integration typically requires full access to mailboxes, so read the permission request before you grant it.
Forward a seeded test message to two colleagues, flag it, then record whether every copy left every mailbox.
Sender authentication is a baseline the platform reads, and your own domain settings decide most of it. To pass DMARC, a message passes SPF or DKIM with alignment, and the domain owner picks the action for failures5. With the reject option, the receiving server rejects the message and usually sends a bounce to the sender.
A reporting mailbox holds what users flag, so the trial shows whether the platform reads those reports and closes them with a verdict.
Confirm your domain's DMARC policy before the trial starts, so the platform is tested on what DMARC misses.
False positives belong on the scorecard next to catches. Microsoft states that false positives rise as more products are added. It also states that no number of layers brings total protection to 100 percent.
A trial that counts only catches rewards the platform that holds the most mail. Use the method in how to evaluate detection accuracy and the miss rate to set both numbers against the same mail.
Log every good message the platform held, alongside every attack it caught, from day one.
Run the trial read-only on live mail first, then score each capability on what it did with your messages. The steps for a read-only test are in how to structure a read-only email security PoC.
| Capability | What to test in the trial | AegisAI |
|---|---|---|
| Intent analysis | Mail your current stack delivered that the platform flags, with its reason | Investigates every inbound email for what it is trying to do |
| Links and files | A seeded link that turns bad after delivery | Follows every link and opens every file before anyone sees the email |
| Payment fraud | A plain text bank change request with no link or file | Catches payment requests even with no link or file to scan |
| Removal after delivery | Every copy of a flagged message pulled, forwards included | One bad email is pulled from every inbox it reached |
| Plain verdicts | An analyst checks the stated reason without opening a console | Every verdict comes with its reason in plain language |
| Sender authentication | Your domain's DMARC policy and the action it sets for failures | No edge. Its page names no DMARC setting; the policy lives at your domain |
| False positives | Good messages held during the trial | No edge. No layer reaches 100 percent |
The last two rows give no edge to any platform, because DMARC is set at your domain and no layer reaches total protection. Fill one column per platform you trial and score each row pass or fail.
AI agents investigate every inbound email for what it is trying to do, and remove the ones that fail from every mailbox they reached. Catches fake executives, vendors and payment requests, even with no link or file to scan. It reads mail through the Microsoft 365 or Google Workspace API after delivery23. One bad email is pulled from every inbox it reached, forwards included. Every verdict comes with its reason in plain language, so your team can check it and share it.
Illustrative example, figures are fictional: Harbor Lane Logistics (fictional) has 1,200 mailboxes and ran a read-only trial for 30 days. The platform flagged 41 messages the current stack had delivered, 9 of them plain text payment requests with no link. A seeded bank change request was flagged within the first hour. A seeded link that turned bad on day 3 was pulled from 4 mailboxes, forwards included. The helpdesk logged 2 good messages held, against 7 a week before the trial. The domain's DMARC policy was already at reject, so the authentication row scored on the domain setting.
Copy the structure and replace every fictional figure with one from your own trial.
Track three numbers during the trial, each in a system you already run.
Put the three weekly figures into the scorecard on day 30 and decide on them.
Eight: intent analysis; link and file analysis; payment fraud detection; sender authentication; removal after delivery; plain verdicts; a user reporting loop; false positive control. Test each one on your own mail in a trial.
Because there is nothing for a link or file scanner to inspect. IC3 recorded $3,046,598,558 in BEC losses in 2025.
Measure it in the trial. Microsoft's zero-hour auto purge searches only the last 48 hours of delivered email, so record whether a candidate pulls a message older than that.
Microsoft says the Graph API integration typically requires granting the service full access to mailboxes. Read the permission request, and check whether removal is granted and revoked separately.
Microsoft states that false positives increase as more products are added. Count good messages held during the trial, next to the attacks caught.
Yes. DMARC tells receiving servers what to do with mail from your domain that fails SPF or DKIM, and that setting lives at your domain, outside any platform.
