How to Tell if Google Chrome Secretly Downloaded a 4GB AI Model to Your Device


Google Chrome could be taking up some extra storage space on your device. Based on reports, the browser has been automatically downloading a 4GB AI model onto some users’ hard drives without their permission. This isn’t the first time Google has discreetly interfered with users’ devices, and privacy advocates say the practice may violate data laws.

The mysterious file in question is Gemini Nano, an AI model that runs on devices, such as smartphones and laptops rather than in the cloud. According to Alexander Hanff, a Swedish computer scientist and lawyer known as That Privacy Guy, it’s been installed on some Chrome browsers without permission. You won’t know when it’s been downloaded onto your device, either. 

Hanff said Gemini Nano will only be installed if the device meets the hardware requirements. It’s still unknown how many people have gotten the install.

Gemini Nano performs tasks such as detecting scam phone calls, helping you write text messages, summarizing recordings and analyzing Pixel phone screenshots. It’s not to be confused with the AI Mode pill in the address bar. If you use AI Mode, your queries are routed to Google Gemini servers, not to Gemini Nano.

AI Atlas

A Google spokesperson told CNET that Gemini Nano will automatically uninstall if the device doesn’t have enough resources, such as processing power, memory, storage space or network bandwidth. 

“In February, we began rolling out the ability for users to easily turn off and remove the model directly in Chrome settings,” the spokesperson said. “Once disabled, the model will no longer download or update.”

Google gives more information about on-device generative AI models in Chrome on this web page.

How to get rid of the AI model 

If you want to remove the 4GB AI model from your device, first check whether it’s installed. 

Hanff said Chrome users will not know they have Gemini Nano unless they search for it, because “Chrome did not ask” and “Chrome does not surface it.”

The easiest way to remove Gemini Nano from your device is to uninstall Chrome.

On a Mac

  1. If you’re using a Mac, open Finder by clicking the blue smiling face icon on the far left of the dock.
  2. Then, click Go in the top menu bar and hold the Option key so that Library appears in the dropdown menu. 
  3. Click Library, then navigate to Application Support > Google > Chrome > Default. See if there’s a folder called OptGuideOnDeviceModel. If the folder exists and contains a file named weights.bin, the AI model was installed.
  4. To permanently remove it on a Mac, open Chrome and click the three-dot menu in the top-right corner. Then click Settings, then System and toggle off On-device AI.

On a Windows device

If you’re running a Windows device, there are a few ways to check whether Gemini Nano is installed.

  1. One way is via a Run Command. Press the Windows key and R, paste in %LOCALAPPDATA%\Google\Chrome\User Data\OptGuideOnDeviceModel and then press Enter. If that file comes up, see if weights.bin is in there.
  2. You can also use File Explorer to check whether the AI model is installed. Navigate to C:\Users\[YourUsername]\AppData\Local\Google\Chrome\User Data\OptGuideOnDeviceModel and look for weights.bin.
  3. To get rid of the AI model in Windows, open Chrome, navigate to Settings > System, and toggle off On-device AI. While still in Chrome, type chrome://flags in the address bar and search for Optimization Guide. Then, set Enables Optimization Guide on Device to Disabled.
  4. Then restart Chrome by completely closing it, using the menu to exit, not just closing windows.
  5. Finally, delete local files by navigating to \AppData\Local\Google\Chrome\User Data and deleting the OptGuideOnDeviceModel folder.

Watch this: Google I/O 2026: New Gemini, Smart Glasses and a Whole New Laptop OS. Here’s What to Expect

Why does it matter?

Hanff said the push might be intended to help Google cut costs by moving AI work off its own servers and onto your computer.

“Running inference on users’ own hardware allows them to push ‘AI features’ without the compute costs,” Hanff told CNET.

AI inference is the process by which the model actually does the things you prompt it to, as opposed to the training of it, which generally happens in a data center. If it’s happening on your computer instead of in the cloud, that could have an impact on things like your computer’s speed or battery life, in addition to storage space the model’s taking up on your hard drive.

But Hanff suggested there could be legal ramifications, at least in Europe. He suggested that the Gemini Nano install could constitute a breach of the European Union’s General Data Protection Regulation’s principles of lawfulness, fairness and transparency. Hanff said that, considering the potential environmental impacts, Google should have announced it under the Corporate Sustainability Reporting Directive.

“Google has given us every reason not to trust them with a history spanning two decades of global privacy violations at massive scale,” Hanff told CNET. “So, I suspect they figured asking permission (what the law requires) would hinder their ability to push this model and, of course, whatever comes after it.”





Source link

Leave a Reply

Subscribe to Our Newsletter

Get our latest articles delivered straight to your inbox. No spam, we promise.

Recent Reviews


Denodo Data Virtualization – Table of Content

Need for Denodo:

Denodo is really the market leader in Data Virtualization, offering data access, data governance, and data delivery capabilities across the most diverse enterprise, cloud, big data, and unorganized information sources without requiring data to be moved from their existing directories.

                       Interested in learning Denodo Join HKR and Learn more on Denodo Training!

Features of Denodo:

Denodo Platform goes above and beyond any other data virtualization solution, providing:

  • A newly designed web-based user interface provides an exceptional user experience that is tailored to the specific needs of business and IT stakeholders.
  • Intelligent and optimal query execution strategy utilizing a Dynamic Query Optimizer for faster data access
  • With assistance of Summaries, Smart Query Acceleration can be used for complex analytical scenarios.
  • In-Memory Parallel Processing accelerates data access to unprecedented speeds.
  • A collection of computerized lifecycle management options that enable users to spend a little less time controlling information and much more time utilizing information to make decisions.
  • A Dynamic Denodo Data Catalog enables seamless data access, improved collaboration, and ML-driven automatic recommendation.
  • Cloud infrastructure management that is automated, with PaaS support for cloud and hybrid environments.
  • A modern data services layer that supports OAuth 2.0, SAML, OpenAPI, OData 4, GraphQL, and other cloud standards for easy integration with existing cloud systems.
  • Denodo can be deployed using marketplaces such as Amazon Web Services (AWS), Microsoft Azure, Google Cloud Platform (GCP), and Docker.
  • Secure, selective availability to a firm’s entire data holdings via a single point of control and management ensures seamless safety and accountability.

The Denodo Platform connects to numerous data sources, combining and publishing data in a business-friendly format to multiple consumers.

Denodo Training

  • Master Your Craft
  • Lifetime LMS & Faculty Access
  • 24/7 online expert support
  • Real-world & Project Based Learning

Data Virtualization:

Data virtualization seems to be a logical data layer which really incorporates all business intelligence silos scattered throughout different systems, maintains the unified data for consolidated security and governance, and provides it in near real – time to corporate customers.

  1. Data virtualization provides a unique strategy to viewing, controlling, and delivering data with no need for a physical repository.
  2. Data virtualization unifies data that has been siloed throughout all enterprise applications, irrespective of data format, destination, or latency.
  3. Data virtualization creates a centrally controlled secure layer for cataloging, searching, discovering, and managing unified data and its relationships.
  4. Data virtualization provides interconnected information about business applicable in real time.

The Real Time Challenge with the Data Virtualization:

Data virtualization is a cutting-edge method for data integration. Unlike ETL solutions, that mostly reproduce data, data virtualization keeps data in source systems while providing data consumers with an accurate understanding of all data.When enterprise customers dig deeper into documents, data virtualization retrieves the information from the resource description systems in real time. Data virtualization demonstrates that accessing the data is far better compared to accumulating it.

Uses of Data Virtualization:

Businesses in a wide range of industries can motivate complex systems with real-time direct connections to comprehensive information.

  • Examine existing business productivity in comparison to previous years.
  • Act in accordance with rules that govern historical information to be traceable.
  • Look for and explore data that is related to one another.
  • Replace legacy applications while modernizing business applications.
  • Simply move from on-premises to cloud-based applications.
  • Data can be digitized by providing it as a service.

Lets’s get started with Denodo Tutorial !

HKR Trainings Logo

Subscribe to our YouTube channel to get new updates..!

How Data Virtualization Works?

To provide a comprehensive view of corporate data to corporate customers throughout all fundamental data sources, data virtualization provides a straightforward three-step process: connect, combine, and consume.

  • Connect to any data source
  • Combine any data type
  • Consume the data in any mode

Connect to any Data Source:

Data virtualization connects to a wide variety of data sources, including databases, data warehouses, cloud applications, big data repositories, and even Excel files.

The Connect layer uses data from diverse repositories while masking the complexities of the upper layers’ underpinning network topologies and formats. Data virtualization makes a connection to a wide variety of data sources, from organised to unorganised, such as databases, big data systems, streaming sources, cloud repositories, the Web, NoSQL sources, and flat files.It accesses particular data repositories or applications using highly specialised connectors and performs data source form conversions and widespread acceptance so all base views occur as relational views to the upper layers.

Combine any type of Data:

Data virtualization integrates pertaining data into actionable views regardless of data forma relational databases, noSQL, Hadoop, web services and Cloud APIs, files, and so on.

The Combine layer efficiently simplifies Web processes by modeling them with a library of pre-built layouts and elements for workflow, connectivity, and retrieval, and also the organizing of Web, semi-structured, and unstructured data. It supports dynamic transitions with logical operators for the smooth development of composite data views on top of the base views provided by the coprocessor.

In this layer, consumers can use SQL and relational techniques those who already are acquainted with to execute large data transformations, metadata designing, quality management, and semantic color coordinated operations.

Consume the Data in any Mode:

Business users can consume data through reports, analysis tools, portals, mobile apps, and Web apps thanks to data virtualization.

The Consume layer provides a unified interface for accessing and interacting with both the inherent data sources, and also abstracted data views in a delivery time format.JDBC, ODBC, ADO.NET, SOAP web services, RESTful web services (output as XML, JSON, HTML, or RSS), OData, portlets and data widgets (JSR-168, JSR-286, or Microsoft Web Parts to be implemented in SharePoint), exports to Microsoft Excel/SQL, and JMS message queues are available in the Consume layer to meet the needs of business users.

Interested in learning Denodo Join HKR Denodo training in Hyderabad!

Denodo Training

Weekday / Weekend Batches

Conclusion:

In this blog post we learned about dendo and how it acts as an most prominent data virtualization tool in today’s global market. Had any doubts drop your quieres in the comments.

Related Artilces:

  1. Denodo Tool
  2. Denodo Architecture
  3. Denodo certification



Source link