Google Chrome and Microsoft Edge are automatically downloading a massive 4GB AI model file to Windows computers, often without users' knowledge or explicit consent. The file, named weights.bin, is part of a local foundational model that powers on-device generative AI features like text prediction and writing assistance. While the intention is to enable faster, privacy-focused AI processing, many users are alarmed by the stealthy download, significant disk space consumption, and potential bandwidth usage.

IT administrators and savvy users can block this behavior using the GenAILocalFoundationalModelSettings enterprise policy via a registry tweak or Group Policy. This article explains why the download happens, its implications, and step-by-step methods to prevent your browser from silently pulling down a 4GB file.

What Is the Local AI Model and Why Is It Being Downloaded?

Google and Microsoft are integrating on-device AI capabilities into their browsers to power features like "Help me write," smart text predictions, and possibly future enhancements such as image generation or translation. These features rely on a local foundational model—a compact version of a large language model (LLM) that runs directly on your hardware without sending data to the cloud.

The model is stored as a file, typically named weights.bin, and is downloaded by Chrome and Edge in the background. The download occurs even if you never explicitly enable or use any AI features. According to reports and user observations, the file can appear in a hidden directory such as %LOCALAPPDATA%\Google\Chrome\User Data\... or a similar Edge profile path.

Which Versions of Chrome and Edge Are Affected?

Based on community reports, this behavior started appearing in Chrome version 121 and later, and in Edge around version 120+. The download is tied to the GenAI component and is enabled by default in many installations. It can weigh between 3.5GB and 4.5GB depending on the specific model version and platform optimizations.

User Impact: Disk Space, Bandwidth, and Privacy

A 4GB download is no small matter, especially for users with limited SSD storage, metered internet connections, or strict data caps. Here’s why this matters:

  • Disk Space Consumption: 4GB is a significant chunk of a 128GB SSD, and the model may be duplicated across multiple browser profiles or both Chrome and Edge if you use both.
  • Bandwidth Usage: The silent download can eat into monthly data limits without any notification. For mobile hotspot users, this can lead to unexpected charges.
  • Performance Overhead: Once downloaded, the model may load into memory when certain browser features are used, consuming hundreds of megabytes of RAM.
  • Privacy Concerns: While the model runs locally, users have not given explicit consent for its download. Some worry about future updates or telemetry associated with the component.

How to Check If the Model Has Been Downloaded

You can verify whether the weights.bin file exists on your system by navigating to your browser’s profile directory:

  • Chrome: %LOCALAPPDATA%\Google\Chrome\User Data\ChromeEnhancedForTesting or a subfolder like OptimizationGuide or GenAILocal.
  • Edge: %LOCALAPPDATA%\Microsoft\Edge\User Data\ and look for similar folders.

Search for files named weights.bin or large folders (around 4GB) created recently. You can also check about:componets (Chrome) or edge://components/ to see if the “GenAI” component is listed and its current version.

Blocking the Download with Enterprise Policy

The most robust method to prevent the download is to disable the GenAILocalFoundationalModelSettings policy. This policy controls whether the browser downloads and uses the local AI model. Setting it to Disallowed (value 0) stops the download entirely.

Method 1: Registry Editor (Manual Tweak)

  1. Press Win + R, type regedit, and press Enter.
  2. Navigate to the appropriate key:
    - For Chrome: HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Google\Chrome
    - For Edge: HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Edge

If the key doesn’t exist, right-click on the parent folder and create it.
3. Right-click in the right pane, select New > DWORD (32-bit) Value.
4. Name it GenAILocalFoundationalModelSettings.
5. Set its value to 0.
6. Restart the browser.

Alternatively, you can apply this to the current user only by replacing HKEY_LOCAL_MACHINE with HKEY_CURRENT_USER.

Method 2: Group Policy (Domain-joined or Pro/Enterprise)

For managed environments, use the Group Policy Editor:

  1. Open gpedit.msc.
  2. Under Computer Configuration > Administrative Templates, locate the appropriate Chrome or Edge policy templates. You may need to download the latest ADMX templates from Google or Microsoft.
  3. Find the policy named “Enable GenAI Local Foundational Model” (or similar) and set it to Disabled.
  4. Update policies with gpupdate /force.

Method 3: Using Command Line Flags (Temporary)

For a non-persistent block, you can launch Chrome or Edge with the following flag:

--disable-features=GenAILocalFoundationalModel

Right-click the browser shortcut, go to Properties, and append the flag to the Target field after the executable path (e.g., "C:\...\chrome.exe" --disable-features=GenAILocalFoundationalModel). Note that this only works when launching from that shortcut and may be overridden by updates.

What About the Already Downloaded Model?

Disabling the policy stops future downloads and prevents the browser from using the model, but it does not automatically delete the existing file. To reclaim space:

  1. Close the browser completely.
  2. Navigate to the profile folder as described above.
  3. Delete the weights.bin file or the entire GenAILocal folder.
  4. Optionally, delete other large AI-related files from the OptimizationGuide or EnhancedForTesting directories.

After deletion, the browser will not re-download the model as long as the policy is in place.

Official Responses and Future Plans

Google has not issued a broad public statement about this download, but it is mentioned in enterprise policy documentation. The feature is part of Chrome’s “Enhanced” functionalities aimed at improving web experiences with on-device AI. Microsoft similarly integrates AI deeply into Edge via Copilot, though the model download mechanism may differ.

There are indications that future Chrome versions could expand the model’s usage to include more features, potentially making the download even larger. For now, the download appears to be a one-time event per installation, but it may be updated periodically.

Should You Block the Model?

Whether to block the download depends on your use case:

  • Block it if: You value disk space, have limited bandwidth, or prefer not to run local AI models for privacy or performance reasons. The features it enables are still experimental and may not be critical.
  • Allow it if: You frequently use AI-powered writing assistance or other on-device intelligent features and want the lowest latency, fully offline operation.

The Bigger Picture: On-Device AI in Browsers

Silent AI model deployments are becoming more common as browsers evolve into application platforms. Chrome and Edge aren’t alone—other software vendors are embedding local models for tasks like speech recognition, OCR, and code completion. The challenge for users and IT administrators is maintaining control over system resources and data.

Microsoft’s approach with Edge is closely aligned with Windows Copilot and integrates with the operating system. Google, lacking a dominant desktop OS, pushes its AI directly into Chrome, which runs on Windows, macOS, and Linux. This cross-platform strategy makes centralized management via policies essential for enterprises.

Community Reaction and Workarounds

On forums like Reddit and tech communities, users have expressed frustration about the undisclosed download. Workarounds beyond the official policy include:

  • Blocking the download domain optimizationguide-pa.googleapis.com or similar endpoints at the network level.
  • Using group policy to restrict component updates.
  • Switching to browser forks that disable such features (e.g., Ungoogled Chromium).

However, these are advanced measures that may break other browser functions.

Conclusion

The 4GB local AI model download in Chrome and Edge is a significant but manageable change. By setting the GenAILocalFoundationalModelSettings policy to Disallowed, you can reclaim control over your storage and bandwidth. As on-device AI becomes the norm, transparency and user consent must be priorities. For now, the registry tweak or Group Policy is your best defense against unwelcome large-scale downloads.

Keep an eye on browser update notes and enterprise policy changes, as Google and Microsoft may introduce more granular controls in the future. In the meantime, regularly audit your browser components and disk usage to stay ahead of silent additions.