Skip to main content
  1. Home
  2. Computing
  3. News

Google quietly rewrote Chrome’s AI documentation with important answers to on-device AI downloads

Google quietly clarified how Chrome handles on-device AI downloads.

Add as a preferred source on Google
Google Chrome with Gemini
Google Chrome

I remember Chrome quietly downloading a 4GB AI model onto PCs. It raised plenty of eyebrows in May 2026. The next day, a Chrome executive clarified how on-device AI is core to its “developer & security strategy,” and that the model processes data locally. 

Now, Chrome has updated its support documents to explain the download and, more importantly, include an option to stop it. 

So what exactly changed in Chrome’s documentation?

Google has quietly updated Chrome’s On-device AI support page between July 29 and August 1 (via WindowsReport). As part of the update, it added four entirely new sections about eligibility requirements, download conditions, guidance for multiple users and devices, and enterprise-specific rules. 

Recommended Videos

As mentioned in the eligibility section, Chrome won’t even attempt to download any AI models unless your device has the right hardware to run them. In simpler words, unsupported or incompatible machines get skipped entirely. 

For devices that do qualify, however, Chrome now says it checks for an unmetered internet connection, along with roughly 20GB of free disk space (including the 4GB AI model). It also analyzes whether the device has enough processing power before downloading anything.

How are you actually supposed to manage this stuff?

Google’s also being clearer about control now. 

The documentation states the “On-device AI” toggle in Chrome’s settings is the only supported way to manage these downloads. It also warns against manually hunting down and deleting model files yourself. Doing that won’t stop Chrome from redownloading them. 

When you switch the toggle off, Chrome removes the downloaded files and disables everything that’s based on them. It only checks eligibility again and redownloads if you flip it back on, but not otherwise. All of this is on a per-user and per-device basis, with IT admins getting separate control through Chrome Enterprise Policy.

This update lands months after people first noticed Chrome silently downloading Gemini Nano, its roughly 4GB on-device AI model, without clearly asking first. Google hasn’t directly connected the dots, but the timing here doesn’t feel coincidental, at least to me.

Shikhar Mehrotra
For more than five years, Shikhar has consistently simplified developments in the field of consumer tech and presented them…
Huawei made its giant folding laptop faster and tougher, but the pen is the real upgrade
Huawei’s giant folding laptop is back tougher, faster, and pen-ready
Huawei MateBook Fold Ultimate Design being used with the M3-Pen

Huawei’s laptop that opens into an 18-inch OLED screen already looked pretty futuristic when it first debuted. But the 2026 update makes the laptop even better. The company has improved performance and display durability, which brings another layer of interaction.

The refreshed Huawei MateBook Fold Ultimate Design unfolds into an 18-inch, 3296 x 2472 dual-layer OLED display with a 4:3 aspect ratio, 1,600 nits of peak brightness, and a 92% screen-to-body ratio. Fold it into a laptop shape, and each half becomes a more portable 13-inch, 3:2 screen. Despite that enormous panel, the device weighs 1.16kg without its detachable keyboard. It measures 7.3mm at its thinnest point when open and 14.9mm when closed.

Read more
Study finds readers rate AI-written stories higher, but still trust the “human” label more
Researchers also found that readers could barely differentiate between AI-generated and human-written stories, though those familiar with AI tools guessed somewhat better.
updated book and AI photo

If you think AI writing is easy to spot, a new study suggests you may be overestimating your ability to tell the difference. Researchers from Villanova University have found that readers struggled to tell AI-generated stories from human-written ones and frequently rated the AI versions higher.

The study, led by Dr. Deena Skolnick Weisberg and published in the journal Judgment and Decision Making (via The Guardian), asked more than 1,600 participants to rate one of six short stories, three written by a human and the rest generated by ChatGPT, on quality and engagement. Participants were told who wrote their story, though that label wasn't always accurate.

Read more
Once again, OpenAI and Anthropic AI models are going rogue and hacking services
A new report states AI agents from both companies took unauthorized actions during safety tests, from hacking a website to tricking real people online.
Claude website open on laptop

OpenAI and Anthropic have both had a rough few weeks on the AI safety front. OpenAI recently disclosed that its models broke out of a test environment and hacked into Hugging Face and four other organizations. The news prompted Anthropic to review its own testing, which revealed that Claude had also gained unauthorized access to three companies.

Now, the UK's AI Security Institute (AISI) has disclosed a new round of incidents (via Wired). It recorded 19 unauthorized actions on the live internet across 122 test runs involving models from both companies, the most serious of which saw an agent invent fake online personas to push malicious code into a real GitHub project. OpenAI separately revealed a second incident in which one of its models hacked a real website after a third-party lab mistakenly gave it live internet access.

Read more