AI and cybersecurity in Microsoft environments: Security Copilot, AI phishing and Copilot adoption
Key takeaways
- Microsoft Security Copilot mainly helps small teams with summaries, guided response and triage in the Defender portal. A monthly allowance is included for Microsoft 365 E5 and E7; the results still need expert review.
- Against AI-generated phishing, rely on technology and processes: phishing-resistant MFA, impersonation protection in Defender for Office 365 and a call-back process for payments.
- Microsoft 365 Copilot only shows what a user is already allowed to open. The risk lies in overly broad permissions; Microsoft Purview makes them visible.
- Management decides who is accountable for AI results and which data is released for AI. IT implements this with Entra ID, Purview and SharePoint.
For companies in Microsoft environments, three questions about AI in cybersecurity matter in practice: what does Microsoft Security Copilot do for your own team? What helps against phishing that no longer stands out linguistically? And what risks arise when staff use Microsoft 365 Copilot? This article answers all three, first for management, then for IT.
What does Microsoft Security Copilot offer a small security team?
Security Copilot shortens the time a team needs to understand an incident and decide. It does not replace an expert, but it speeds up routine work and lowers the barrier for less experienced team members.
In the Defender portal, Security Copilot summarises incidents, suggests response steps, analyses suspicious scripts and files, and creates incident reports. In addition, there are agents, for example for triaging phishing reports submitted by staff. Agents have to be set up deliberately; they are not active automatically.
On licensing: Security Copilot is included for Microsoft 365 E5 and E7, with 400 Security Compute Units (SCUs) per month for every 1,000 paid user licences, up to 10,000 SCUs per month. The allowance also scales below 1,000 licences. The rollout has been proceeding in phases since November 2025; the message center in the Microsoft 365 admin center shows whether your tenant has been enabled. Without E5 or E7, Security Copilot is billed through provisioned SCUs under the existing pricing model.
Important for IT: Security Copilot processes and stores data from Microsoft 365 services. After automatic provisioning, review the owner settings, the preselected roles and the data storage location, and define that AI suggestions are confirmed by an accountable person before any intervention in operations.
What helps against AI-generated phishing?
Flawless language is no longer a warning sign, so defence must not depend on the attentiveness of individual employees. Three complementary layers are effective.
Phishing-resistant MFA. Classic MFA with push approval or a code can be intercepted through a convincing fake sign-in page. Phishing-resistant methods are bound to the genuine sign-in address: passkeys (FIDO2), Windows Hello for Business and certificate-based authentication. In Microsoft Entra Conditional Access, you enforce this through the “Phishing-resistant MFA” authentication strength, first for administrators and particularly exposed roles. How such attacks on Microsoft 365 work in principle is shown in our webinar recording.
Impersonation protection in Defender for Office 365. SPF, DKIM and DMARC prevent someone from spoofing your own domain. They do not help against lookalike domains, because such messages pass the checks. The anti-phishing policies in Defender for Office 365 detect impersonation of users and domains. In the default policy, this protection is not fully configured: add management, finance and important partner domains as protected senders and set the action to quarantine.
Approval processes. No payment and no change of bank details based solely on an email, a call or a voice message. A four-eyes principle and a call-back to known contact details work even when a voice sounds deceptively real.
What risks arise when introducing Microsoft 365 Copilot?
Microsoft 365 Copilot only shows content that a user is at least allowed to read. That sounds reassuring, but it shifts the risk to your permissions: a SharePoint site that has accidentally been shared with all employees suddenly becomes easy to find through Copilot. For management, the order is therefore decisive: clean up permissions first, then roll out Copilot widely.
For IT:
- Find oversharing: Data Security Posture Management for AI in Microsoft Purview runs a weekly data risk assessment for the most-used SharePoint sites. SharePoint Advanced Management reports additionally show sharing with “Everyone except external users”.
- Label what is confidential: Copilot respects sensitivity labels with encryption. DLP policies for the Microsoft 365 Copilot location prevent labelled content from being processed in responses.
- Ensure traceability: Auditing in Purview must be active so that AI interactions can be evaluated.
- Control non-Microsoft AI services: Global Secure Access lets you define which external AI tools can be reached.
How an auditable rollout works in practice is described in the articles AI adoption with Entra ID, Purview and SharePoint and Controlling AI tools with Global Secure Access.
What should management and IT decide now?
- Assign accountability: Who confirms AI suggestions in security operations, and who releases data areas for Copilot?
- Phishing-resistant sign-in for administrators: This is one of the most effective measures against stolen access.
- Configure impersonation protection: Add protected users and domains in Defender for Office 365.
- Review permissions before the Copilot rollout: Evaluate the data risk assessment in Purview and close oversharing.
The next step
We make AI adoption audit-ready. In the AI Foundation, we prepare permissions, sensitivity labels and policies in Entra ID, Purview and SharePoint so that Microsoft 365 Copilot and other AI services can be introduced in a traceable way, with evidence.
Frequently Asked Questions
Is Security Copilot active automatically with Microsoft 365 E5?
Eligible E5 and E7 tenants receive Security Copilot step by step through automatic provisioning, without their own Azure setup. You still have to set up agents yourself, and you should review the privacy and role settings after provisioning.
Are SPF, DKIM and DMARC enough against phishing?
No. These standards prevent spoofing of your own domain. Messages from lookalike domains, however, pass the checks. Impersonation protection in Defender for Office 365 helps against this, and phishing-resistant MFA helps if someone does end up on a fake sign-in page.
Can Microsoft 365 Copilot show data that employees are not allowed to see?
Copilot only shows content that the respective user has at least read access to. The problem arises when permissions are too broad. That is why a review of sharing and sensitivity labels belongs before the broad rollout.
Does Security Copilot replace experts in security operations?
No. Security Copilot speeds up analysis and documentation, but decisions about interventions in operations are still made by an accountable person. Without defined ownership, even a good AI suggestion goes unused.