What the add-on is, and what it isn't
Microsoft Learn describes it as an optional layer on Azure Firewall Premium. It doesn't replace the Basic, Standard, or Premium SKUs. Premium already provides TLS inspection and signature-based intrusion detection and prevention. AI Security is a separately enabled addition and doesn't change those features.
Azure's "In preview" label means the feature is for non-production use and testing. Microsoft hasn't published a general availability date, performance figures, detection-accuracy numbers, or a service-level commitment for this add-on.
The five capabilities
| Capability | What it does | Boundary to note |
|---|---|---|
| Prompt Threat Protection | Detects direct and indirect prompt-injection attempts in supported AI traffic, with an option to block. Microsoft's announcement says it uses Azure AI Content Safety Prompt Shields and needs no per-application SDK integration. | Needs an Azure AI Content Safety resource, which the firewall policy reaches through a managed identity. |
| Network DLP | Uses Microsoft Purview classification and policy decisions to stop sensitive data going to unsanctioned destinations, including AI apps and models. | Needs Purview policy configuration and a user context. Purview policy scope affects the result. |
| Advanced Malware Threat Protection | Analyzes supported files in scoped application traffic, with alert or deny thresholds. The announcement says it uses small language models tuned for network security that run on CPUs. | Covers only supported files and selected traffic. |
| Advanced Application Threat Protection | Analyzes eligible application requests for exploit attempts, with alert or deny thresholds. | Covers only eligible requests under the applied configuration. |
| Policy Advisor | Analyzes policy configuration and traffic logs to produce findings with supporting evidence. Examples include unused rules, rule ordering, and traffic anomalies. | It requires Policy Analytics. It doesn't change policies automatically, and Microsoft recommends a dry-run before you apply changes. |
Enabling it is not the same as inspecting traffic
This is the main point in Microsoft's documentation. Turning the add-on on doesn't show that any particular flow was inspected. Traffic has to match the applicable rules and the protection configuration.
Microsoft organizes the configuration like this:
- A security profile holds the settings for one protection capability.
- A security profile group combines profiles and is associated with application rules in Azure Firewall Policy. A group can hold at most one profile of each type.
- The application rule selects the traffic. The group decides how that traffic is inspected and what action applies.
Azure Firewall processes network rules before application rules. A matching network rule can therefore keep a flow from ever reaching the application rule that carries your security profile group. An allowed request is not evidence that Prompt Threat Protection looked at it.
Deployment checklist for the preview
- Check the region. The preview is available in East Asia, UK South, and West Central US. "Public preview" does not mean every region.
- Plan downtime. Enabling the add-on on an existing firewall requires downtime, so schedule it as maintenance.
- Enable it on every firewall. If one policy is associated with several firewalls, enable the add-on on each of them before you configure the protections.
- Use the Azure portal. During the preview you configure the add-on and its protections there. Azure PowerShell, Azure CLI, Terraform, and the REST API aren't supported for this configuration yet. That is a real obstacle for teams that manage infrastructure as code.
- Get TLS inspection working. For HTTPS traffic the firewall must decrypt successfully, and clients must trust the inspection certificate authority. The add-on doesn't remove that requirement.
- Line up the dependencies. These are the Content Safety resource with its managed identity, Purview policies, and Policy Analytics, depending on which capabilities you use.
- Don't assume coverage. Selecting one protection doesn't turn on the others. Coverage also varies by protocol, content type, and traffic direction.
An SDK-level detail is also visible. The Azure .NET management library's prerelease package has a property for whether the AI security add-on is enabled on a firewall. That fits Microsoft's statement that the dedicated preview workflow isn't supported through programmatic tooling yet. A property existing doesn't mean the preview configuration is supported through code.
Context and analysis
The Network DLP piece matches an earlier roadmap item. WindowsForum previously covered a Microsoft roadmap entry for Purview DLP integration with Azure Firewall. That entry targeted a public preview in November 2026 and general availability in March 2027. The roadmap item is separate from this announcement, and the preview documentation gives no general availability date. Treat March 2027 as a roadmap target only. The roadmap coverage also said licensing for the Azure Firewall integration had not been disclosed.
Licensing is still thin. Microsoft says AI Security is licensed as an add-on to Premium and sends readers to Azure Firewall pricing. I haven't verified a price, so check the pricing page before you budget.
My own view is that a network-layer control for AI traffic is attractive. It sits in the path and needs no changes to application code. The cost is operational complexity. You now have to get rule order, certificate trust, profile group scoping, and cross-service identity right before you can say the control is enforced. In preview, with limited regions and portal-only setup, it suits a lab or a pilot on non-production traffic. Test it with deliberate positive and negative cases, for example a known prompt-injection string and a labeled sensitive file. Then confirm in the logs that each was evaluated, rather than assuming so.
Bottom line
- AI Security is an optional Premium add-on, in public preview only.
- It offers prompt-injection protection, Purview-backed network DLP, malware and application-request analysis, and an advisory-only Policy Advisor.
- Preview limits are three regions, portal-only configuration, and downtime on existing firewalls.
- Inspection happens only when traffic is routed, TLS-inspected, and matched to the right application rule. Network rules are evaluated first.
References
- (In preview) Public Preview: AI Security add-on for Azure Firewall Premium Azure Updates · 2026-11-17T20:27:29Z
- Azure Firewall AI Security concepts (preview) | Microsoft Learn learn.microsoft.com
- Public Preview: AI Security add-on for Azure Firewall Premium — Cloud Changelog cloudchangelog.com