Stop Chrome 147 From Secretly Downloading 4GB AI Model

Google Chrome 147 has begun silently downloading a roughly 4GB on-device AI model to users' computers — without asking permission, providing notification, or offering an opt-out toggle. The file, called weights.bin, supports Google's Gemini Nano local AI features, and it reinstalls itself automatically even after users manually delete it.

The issue, first flagged by tech outlet NeoWin on May 7, 2025, has sparked frustration among users who discovered gigabytes of storage consumed without explanation. Fortunately, a straightforward Windows registry edit can permanently block the download.

Key Facts at a Glance

• Chrome 147 silently downloads a ~4GB AI model file called weights.bin
• The model powers Gemini Nano, Google's on-device AI engine
• No user consent is requested, and no opt-out switch exists in Chrome settings
• Deleting the file manually does not work — Chrome re-downloads it automatically
• A Windows Registry modification can block the download on both Chrome and Microsoft Edge
• The fix works on Windows 11 and is also applicable to Windows 10 systems

What Google Chrome Is Downloading — and Why

Google has been aggressively pushing on-device AI capabilities into Chrome as part of its broader strategy to embed Gemini across its product ecosystem. Gemini Nano is the smallest variant in Google's Gemini model family, designed specifically to run locally on consumer hardware without requiring a cloud connection.

The on-device model enables features like real-time text summarization, smart compose suggestions, and local language processing directly within the browser. Unlike cloud-based AI calls to services like Gemini Pro or Gemini Ultra, Nano processes data entirely on the user's machine — which Google frames as a privacy benefit.

However, the delivery mechanism has proven deeply controversial. The weights.bin file — approximately 4GB in size — downloads in the background during normal Chrome usage. Users receive no notification, no progress bar, and no prompt asking whether they want the feature. For users on metered connections, limited storage devices, or older hardware, this silent download represents a significant and unwelcome imposition.

Why This Has Angered Users and Privacy Advocates

The backlash centers on 3 core issues that touch on trust, transparency, and user autonomy.

First, consent is entirely absent. Modern software norms — especially in Europe under GDPR principles and in the US under evolving digital rights frameworks — increasingly emphasize informed user choice. Silently installing a 4GB component violates those expectations.

Second, persistence is aggressive. Users who discover the file and delete it find that Chrome simply re-downloads it. This behavior mirrors patterns typically associated with unwanted software rather than a mainstream browser used by over 3 billion people worldwide.

Third, there is no settings toggle. Unlike other experimental Chrome features that can be managed via chrome://flags or the settings panel, Gemini Nano's model download offers no user-facing control. The only reliable method to stop it requires editing the Windows Registry — a step most casual users would never attempt.

Critics have drawn comparisons to Microsoft's controversial AI Recall feature in Windows 11, which similarly faced backlash for capturing user activity without adequate consent mechanisms. Both cases illustrate a troubling industry pattern: shipping AI features first and addressing user concerns later.

Step-by-Step: Block Chrome's AI Model Download via Registry

The following registry modification prevents Chrome from downloading the Gemini Nano model. This method works on Windows 11 and should also function on Windows 10.

Step 1: Open the Registry Editor

Click the Start Menu and type 'regedit' or 'Registry Editor' into the search bar. Right-click the result and select 'Run as administrator.' Confirm the User Account Control prompt.

Step 2: Navigate to the Chrome Policy Key

In the Registry Editor, navigate to the following path:

HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Google\Chrome\

If the Google or Chrome folders do not exist under the Policies key, you will need to create them manually. Right-click on Policies, select New > Key, name it 'Google,' then create another key inside it named 'Chrome.'

Step 3: Create and Set the Registry Value

Inside the Chrome key, right-click in the right pane and select New > DWORD (32-bit) Value. Name it:

GenAILocalFoundationalModelSettings

Double-click the newly created value and set its data to 1. Click OK and close the Registry Editor.

Step 4: Restart Chrome

Close all Chrome windows completely and relaunch the browser. The registry policy will now prevent Chrome from downloading the Gemini Nano model files.

Bonus: Block the Same Behavior in Microsoft Edge

Microsoft Edge, which shares Chromium's codebase, may exhibit similar AI model download behavior. Users can apply an analogous fix:

• Navigate to HKEY_LOCAL_MACHINE\SOFTWARE\Policies\Microsoft\Edge\
• Create a DWORD value named GenAILocalFoundationalModelSettings
• Set the value to 1

This preemptively blocks Edge from downloading comparable on-device AI components.

Industry Context: The Rush to Ship On-Device AI

Google's aggressive push reflects a broader industry race to bring AI capabilities directly to user devices. Apple integrated its Apple Intelligence suite into iOS 18 and macOS Sequoia with on-device models. Microsoft embedded Copilot features deep into Windows 11 with dedicated NPU support via Copilot+ PCs. Qualcomm, Intel, and AMD are all shipping processors with dedicated AI accelerators specifically to power these local models.

The strategic rationale is clear:

• Reduced cloud costs — every query processed locally saves server resources
• Lower latency — on-device inference avoids round-trip network delays
• Privacy positioning — local processing keeps data on the user's machine
• Offline capability — features work without an internet connection
• Competitive pressure — no major browser maker wants to fall behind on AI

Yet the Chrome incident highlights a critical tension. Companies are so eager to deploy AI features that they skip basic UX principles around user consent. The 4GB silent download is not a bug — it is a deliberate product decision that prioritizes feature deployment speed over user agency.

What This Means for Different Users

Everyday consumers with limited storage — particularly those using laptops with 128GB or 256GB SSDs — may notice meaningful storage loss. A 4GB download represents over 3% of usable space on a 128GB drive.

Enterprise IT administrators face a different challenge. In managed environments with hundreds or thousands of machines, uncontrolled 4GB downloads per device can strain network bandwidth and violate software deployment policies. The registry fix can be deployed at scale via Group Policy Objects (GPO) or endpoint management tools like Microsoft Intune.

Developers and power users should be aware that blocking the model will disable any Chrome features that rely on Gemini Nano for local inference. Currently, these features are limited, but Google is expected to expand on-device AI capabilities significantly throughout 2025.

Looking Ahead: Will Google Add a Proper Opt-Out?

The pressure on Google to provide a user-facing toggle is mounting. Community forums, Reddit threads, and tech publications have universally criticized the silent download approach. Google has not yet issued an official response addressing the consent concerns.

Historically, Google has responded to similar backlash by eventually adding controls — the company reversed course on several controversial Chrome privacy changes after public outcry. A settings-level opt-out for Gemini Nano downloads would be the minimum acceptable response.

Until then, the registry modification remains the only reliable method to prevent Chrome from consuming storage and bandwidth for AI features many users neither want nor need. Users who value control over their systems should apply the fix immediately — and monitor future Chrome updates for any changes to this behavior.

The broader lesson is clear: as AI becomes embedded in every layer of consumer software, the industry must establish clearer norms around consent for large, resource-intensive AI components. Shipping a 4GB model silently is not innovation — it is overreach.