Windows 12 Security Features: How to Configure AI-Driven Local Threat Detection

There is no confirmed Windows 12 security feature or universal Windows 12 setting for AI-driven local threat detection. Microsoft has not announced an official Windows 12 release date, and its current public documentation continues to focus on Windows 11. Avoid unofficial websites offering “Windows 12” downloads.

The closest confirmed technology is AI-agent runtime protection in Microsoft Defender for Endpoint. This public-preview enterprise feature monitors supported local AI agents on managed Windows devices for prompt-injection attacks. It can record suspicious activity in Audit mode or block supported activity in Block mode. It is not a standard consumer option in Windows Security and should not be presented as a confirmed Windows 12 feature.

What the Protection Does

Local AI agents can operate with a user’s permissions. Depending on the agent and its configuration, they may read files, invoke tools, access websites, call services or run commands. A prompt-injection attack places malicious instructions inside a webpage, repository, document or tool response, potentially causing the agent to take an unsafe action.

Defender can inspect activity at three points:

The user’s prompt before the agent processes it.
A tool request before the tool runs.
A tool response before it returns to the agent’s reasoning process.

Microsoft documents two inspection methods. Agent-native event inspection uses vendor-supported hooks exposed by the agent. Network inspection examines supported traffic between an agent and a large language model when native hooks are unavailable. Coverage depends on the agent and its integration.

The documented agents with native event-inspection support include Claude Code, Codex CLI, GitHub Copilot CLI and the GitHub Copilot app. Preview coverage may change, so administrators should check Microsoft’s current support list before deployment. Discovery of an agent does not necessarily mean that runtime inspection or blocking is available for it.

What You Need Before Starting

This is an enterprise Defender for Endpoint capability, not a general Windows 11 consumer setting. Before configuring it, an organization needs an eligible Microsoft security license, devices onboarded to Microsoft Defender for Endpoint, Microsoft Defender Antivirus running in active mode, real-time protection enabled, current Defender platform and engine updates, and at least one supported local AI agent.

Administrators also need suitable permissions to manage Intune or Defender endpoint-security policies and investigate alerts. During the public preview, Microsoft recommends using test devices that receive Defender platform and engine updates through the Beta Channel.

For a single device, open PowerShell as administrator. Check the antivirus signature version with:

Get-MpComputerStatus | Select-Object AntivirusSignatureVersion

Microsoft’s setup documentation lists 1.451.224.0 or later at the time of publication. Preview requirements can change, so verify the current documentation before deployment.

Choose Audit or Block Mode

Mode	What happens
Disabled	Defender does not inspect the selected agent activity.
Audit	Defender detects and records suspected prompt injection but allows the agent action to continue.
Block	Defender blocks supported malicious activity before the relevant action proceeds.

Microsoft recommends beginning with Audit mode on a small group of test devices. Review detections and possible false positives before moving selected device groups to Block mode.

Configure a Single Device with PowerShell

1. Enable agent-native inspection

For agents that expose supported event interfaces, run:

Set-MpPreference -AiAgentProtection Audit

After testing, replace Audit with Block when active enforcement is appropriate. To disable the method, use:

Set-MpPreference -AiAgentProtection Disabled

2. Enable network inspection when required

For supported agents without native event interfaces, run:

Set-MpPreference -AiAgentNetworkInspection Audit

After validation, change Audit to Block. To disable network inspection, use:

Set-MpPreference -AiAgentNetworkInspection Disabled

Network inspection is not universal. Unsupported protocols and protected traffic paths may remain outside its coverage.

3. Verify the settings

Use this command to check the effective configuration:

Get-MpPreference | Select-Object AiAgentProtection, AiAgentNetworkInspection

Close PowerShell and any terminal windows used to run the agent. Open a new terminal before starting the agent again so the supported inspection hooks are applied correctly.

Deploy the Policy with Microsoft Intune

For centrally managed Windows devices, open the Intune admin center and go to Endpoint security → Antivirus. Create a Windows policy and select the Microsoft Defender AI agent runtime protection profile.

Configure Ai Agent Protection as Audit for a limited device group. Review detections, expand the audit deployment, and move suitable groups to Block only after the results are understood.

The Intune profile configures agent-native event inspection. To deploy network inspection through Intune, use a PowerShell platform script containing:

Set-MpPreference -AiAgentNetworkInspection Audit

Replace Audit with Block when enforcement is approved. The script must run in the system context so it can change Microsoft Defender Antivirus preferences.

Test and Review Detections

Use Microsoft’s official benign demonstration to verify the configuration. Testing requires a supported local AI agent, Audit or Block mode, current Defender updates, newly opened agent and terminal sessions, and authorization to inspect the organization’s alerts.

In Block mode, Defender can display a message in the agent terminal, show a Windows toast notification, record the event in Windows Security → Virus & threat protection → Protection history, and create a Suspicious AI prompt injection alert. In Audit mode, the action continues but the detection is recorded.

Administrators can investigate alerts through the Microsoft Defender portal, device timeline and correlated incidents. Advanced Hunting can be used to query related activity with Kusto Query Language.

Common Problems and Fixes

The setting is missing

Check Defender for Endpoint onboarding, licensing, active-mode Microsoft Defender Antivirus, real-time protection, current updates and the supported-agent list. The feature is not expected to appear as a general Windows Security option on a personal Windows 11 computer.

The agent refuses the test but Defender shows no alert

The agent may have rejected the prompt using its own safety controls. That does not prove Defender detected it. Check the runtime setting, update status, active mode and supported-agent list, then close old sessions and start a new one.

Block mode interrupts legitimate work

Return the affected device group to Audit mode while the security team reviews the alert. Preview coverage and false positives may change as Microsoft expands support.

Discovery finds an agent, but blocking is unavailable

Local-agent discovery and runtime protection are separate capabilities. Defender may inventory an agent, its version, host process, user, device, trust status, auto-approval setting and MCP servers without providing runtime blocking for that agent.

“Local” does not mean completely offline

The AI agent may run on the endpoint, but Defender alerts, incidents, portal investigation and threat-intelligence workflows can involve organizational services. This is not an entirely offline security system.

How It Differs from Other Windows Security Features

Smart App Control is a separate Windows 11 feature that evaluates applications before they run. AI-agent runtime protection monitors supported prompts, tool requests and tool responses during agent operation. Microsoft Defender Antivirus and endpoint detection and response provide broader protection against malware and suspicious endpoint activity.

Recall is also different. It is an opt-in Windows 11 feature for Copilot+ PCs that saves and searches local screen snapshots. It is designed for information retrieval, not prompt-injection or malware detection.

Practical Advice for Users and Organizations

Keep Windows 11, Microsoft Defender and security intelligence updates current.
Treat AI agents as software operating with the user’s permissions.
Limit file, shell, browser and network access to what the agent requires.
Review auto-approval settings and connected MCP servers.
Use Audit mode before enabling Block mode across an organization.
Combine runtime protection with least privilege, application control, network restrictions and normal security monitoring.
Do not install purported Windows 12 software from unofficial websites.

Frequently Asked Questions

Is Windows 12 available now?

No official Microsoft source confirms a Windows 12 release date or Windows 12-specific security architecture. Current Microsoft information continues to focus on Windows 11.

Can a normal Windows 11 user enable this protection?

Not as a universal consumer Windows Security switch. The confirmed feature is Microsoft Defender for Endpoint AI-agent runtime protection, which requires managed devices, enterprise infrastructure, supported agents and appropriate licensing.

Does it detect ordinary malware?

Its documented focus is prompt injection and unsafe activity involving supported local AI agents. It complements Microsoft Defender Antivirus, Smart App Control, application control and endpoint detection and response.

What is the difference between Audit and Block?

Audit records and reports suspected prompt injection while allowing the action to continue. Block stops supported malicious activity before execution and can notify the user.

Is Recall the same as local AI threat detection?

No. Recall is an opt-in local snapshot-search feature for Copilot+ PCs. It is intended for information retrieval, not prompt-injection or malware detection.

What should Bangladeshi organizations know?

No reliable source confirms Bangladesh-specific pricing, rollout dates or consumer availability. Organizations in Bangladesh can assess the technology if they use supported Microsoft-managed endpoints, Intune and Defender for Endpoint. Deployment depends on licensing, tenant configuration, device onboarding and supported integrations rather than geography alone.

Conclusion

There is no confirmed Windows 12 security setting to configure today. The practical documented technology is Microsoft Defender for Endpoint’s preview AI-agent runtime protection for managed Windows environments.

Organizations should begin with supported test devices and Audit mode, review detections through Windows Security and the Defender portal, and move selected groups to Block mode only after validation. Individual users should keep Windows 11 and Defender updated, limit AI-agent permissions, review connected tools and avoid unofficial “Windows 12” downloads.

The protection addresses a specific risk: malicious instructions that manipulate AI agents into unsafe actions. It is limited to supported agents and inspection paths and should be used alongside antivirus, application control, least privilege and security monitoring.

Search This Blog

TechBD Daily