AegisAI Perspectives · 144
AegisAI reads each email through the API after delivery, checks it in parallel, sandboxes its links and files, gives one verdict, and removes what fails.
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.

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 reached1. The steps below follow one email from arrival to removal.
If you run IT for a company of 20 to 500 people on Microsoft 365 or Google Workspace, with a small IT or security team and no SOC, this page shows what happens when a tool investigates one suspicious email, and what evidence you get back. The product side is on the email phishing prevention page. What a team of agents is, as a category, is in multi-agent AI email security. How to judge a tool that investigates is in how to pick AI agents that investigate phishing emails. One evasive link, start to finish, is in the evasive links and attachments post. This page is the full path of one message.
Key takeaways
- The message is read through the Microsoft 365 or Google Workspace API after delivery2.
- Four checks run on each message at the same time3.
- Links and attachments are opened in a sandbox4.
- Every verdict comes with a confidence score and plain-language evidence5.
- Convicted mail is pulled from every mailbox, forwards included6. Nothing is removed until you grant write access7.
| Question | Answer | Source |
|---|---|---|
| How does AegisAI investigate an email, step by step? | It reads the message through the API after delivery and runs four checks in parallel. It opens links and attachments in a sandbox. It gives one verdict with its evidence, then pulls convicted mail | Platform overview, aegisai.com |
| What evidence does a verdict carry? | A confidence score and plain-language evidence. For a link, the redirect chain is recorded in the verdict10. An alert adds the warning signs found and a recommended next step15 | Three aegisai.com pages |
| How long does a verdict take? | AegisAI reports on its inbound email security page that the median verdict takes 1.9 seconds, and a convicted message is removed from every mailbox it reached within seconds8 | Inbound email security page, aegisai.com |
| How is a convicted email removed? | It is pulled from every mailbox it reached, once you have granted write access | Platform overview, aegisai.com |
| What API access does moving a message need? | On Microsoft 365, the Graph message move call lists Mail.ReadWrite as its least privileged permission17. In Gmail, the trash call takes the mail.google.com or gmail.modify scope20 | Microsoft Learn, Google for Developers |
Read the rows in order: they follow the email from arrival to removal.
Short on time? Watch the explainer, or book a demo.
Email investigation agents are separate checks that read the same message at the same time. There are four: reputation, intent, policy and headers. Links and attachments are opened in a sandbox. Each check feeds one verdict.
Intent matters most when a message carries nothing to detonate (open in a sandbox to see what it does). Business email compromise is a scam aimed at businesses and individuals who perform a transfer of funds25. Such a message often has no link and no file. Intent is one of the four checks that run on every message.
Ask for the intent evidence first when a payment or bank detail change is flagged.
The email is delivered to the mailbox. AegisAI reads mail through the Microsoft 365 or Google Workspace API after delivery.
The mail platform has already stamped its authentication results. The email authentication results (SPF, DKIM, DMARC) are recorded in the Authentication-results header of inbound messages24.
Four checks start at once: reputation, intent, policy and headers.
Links are opened in a sandbox. Each link is followed hop by hop in an isolated browser, and the chain is recorded in the verdict.
Per the evasive links and attachments post, attachments of all kinds are pulled apart in secure, sandboxed environments14.
The checks end in one verdict, with its score and its evidence.
The platform overview states that convicted mail is pulled from every mailbox, forwards included. This step needs write access.
Keep this order in mind when you read a verdict: each piece of evidence comes from one of these steps.
| Step | What is checked | Evidence recorded | Source |
|---|---|---|---|
| Read after delivery | The delivered message, through the Microsoft 365 or Google Workspace API | The delivered message, read through the API | |
| Authentication | SPF, DKIM and DMARC, checked by the mail platform itself | The Authentication-results header, stamped by the platform for every reader alike | |
| Four checks in parallel | Reputation, intent, policy, headers | Each check feeds the one verdict | |
| Links | Every link, followed hop by hop in an isolated browser | The redirect chain, recorded in the verdict | |
| Gated pages | A page hidden behind a gate | The gate itself, treated as evidence | 11 |
| QR codes | A QR code in the message | Decoded to a URL and detonated like any other link | 12 |
| Attachments | Attachments of all kinds | Pulled apart in secure, sandboxed environments | |
| Verdict | Everything above | A confidence score and plain-language evidence | |
| Alert | The convicted message | Warning signs found, with a recommended next step | |
| Removal | Every mailbox the message reached, forwards included | Runs only after write access is granted |
The authentication row gives no reader an edge: the platform stamps the same header for every tool that reads the message. Use the evidence column as your checklist when you review a verdict.
The email phishing prevention page states that AegisAI treats the gate itself as evidence and never scores a hidden page clean. A QR code is decoded to a URL and detonated like any other link. The redirect chain of each link is recorded in the verdict.
The case of one evasive link is in the evasive links and attachments post. Open the redirect chain in the verdict before you release a gated link.
The integration is an API connection to Microsoft 365 or Google Workspace. Removal waits for write access, and each verdict shows its evidence, so a small team can review it.
Write access is what lets a tool move mail. On Microsoft 365, the Graph move call creates a new copy of the message in the destination folder and removes the original18. Its least privileged permission is Mail.ReadWrite. In Gmail, the trash call moves the specified message to the trash. The gmail.modify scope does not allow immediate, permanent deletion of threads and messages, bypassing the trash19.
Grant read access first, review the verdicts, then grant write access.
The difference shows in three answers: when the tool reads the message, what evidence it returns, and whether it can remove mail after delivery. Ask any detection tool, legacy rules or agents, the same three questions. When does it read the message? What evidence does it return? Can it remove mail after delivery? For this workflow the answers are: after delivery, through the API; a scored verdict with its evidence; and removal from every mailbox it reached.
Microsoft 365 also acts after delivery. Zero-hour auto purge (ZAP) retroactively detects and neutralizes malicious phishing, spam or malware messages delivered to cloud mailboxes23. Automated investigation in Defender for Office 365 Plan 2 analyzes the alert and the message, with the evidence around it21. Its Evidence tab shows the entities analyzed and the results22.
Put the three questions to every tool that investigates your mail, including the ones already in your tenant.
The trail is the verdict itself: its score and its evidence. For a link, it holds the redirect chain. The alert adds the warning signs found and a recommended next step.
This page names no regulation or standard for this record. Ask your auditor which fields they need, then check them against one real verdict.
AegisAI reads mail through the Microsoft 365 or Google Workspace API after delivery. Reputation, intent, policy and headers are checked in parallel. Every verdict comes with a confidence score and plain-language evidence. Convicted mail is pulled from every mailbox, forwards included.
Run it on your own mail. In monitoring mode the agents investigate live mail and remove nothing, while your team reads each verdict. To set that up, ask to start in monitoring mode on your own tenant.
Pick three recent suspicious emails and read their verdicts side by side.
Track three numbers from your own verdicts and alerts.
Compare the three numbers on day 1 and day 30.
AegisAI reads mail through the Microsoft 365 or Google Workspace API after delivery. Four checks run in parallel. Links and attachments are opened in a sandbox. One verdict follows, with its evidence. Convicted mail is pulled from every mailbox, forwards included.
Yes, hop by hop in an isolated browser, and the chain is recorded in the verdict.
The QR code is decoded to a URL and detonated like any other link.
No. Nothing is removed until you grant write access. Once it is granted, convicted mail is pulled from every mailbox, forwards included.
Alerts give a summary, the warning signs found and a recommended next step. The verdict behind the alert holds the evidence.
Not in one step. The Graph move call creates a new copy in the destination folder and removes the original. The Gmail trash call moves the message to the trash, and gmail.modify does not allow immediate, permanent deletion.