Chrome's New AI Features

Another showcase of Google using their dominant market position for Chrome to gain advantage in other markets, like AI agents.

So they integrate Gemini to summarize open web pages and consolidate all your open tabs into summaries. (Open lot's of pages, then summarize them all.) You can search your history with natural language and type Gemini queries directly into the address bar.

This will give them a cognitive profile of you: reading comprehension, decision-making patterns, knowledge gaps, etc.

Scary.

This sounds an awful lot like Microsoft's Recall, only implemented in the browser.

Granted, there have been a lot of times I have trouble finding a website in my history, open tabs or even bookmarks, so I could potentially see how that might be advantageous as long as I was in a situation where I had a second browser for "non-work" related tasks, or this was strictly prohibited in in-private mode.

>4. Find webpages you previously visited

>For those frustrating instances when you want to jump back into a past project but don’t want to scroll through your history to find an important website you previously visited, soon you’ll be able to use Gemini in Chrome to recall it for you. Once launched, you can try prompts like “what was the website that I saw the walnut desk on last week?” or “what was that blog I read on back to school shopping?”

As for their "agentic browsing assistant", I don't have much trouble adding stuff to my shopping cart or other minor tasks. I'm still waiting on that 'Google Duplex' [1] feature they announced years ago that claimed Google would make phone calls for me to make appointments and etc. Make a doctor's appointment? Dispute a charge on my bill? That's what I want.

[1] https://youtu.be/D5VN56jQMWM

Sometimes I wish companies would stop forcing AI features down our throats and putting them just everywhere. At least I hope I can properly disable all of this. I don't need an AI agent scanning everything behind my back.

Some of these features would be nice to have, but I'm not sure I even want my browser to have these capabilities unless there is a mechanism to keep it all local. This is a monumental change to the amount of user data Google can collect via Chrome.

This will surely bring the users back after killing adblockers /s

More AI spyware running on user devices?

Are you kidding me!? We are living in the future and the first thing we (have to) worry about is that something will be used against us.

I really want the 2008 Google where everything they made was welcomed and not hated on sight.

Agentic browser? This. is. what. I. want.

Asking the browser about "that specific thing I might have seen last week?" Sign me the f up!

I'm not being sarcastic, I really wish I could have all of this and not having to worry about antagonistic companies and governments.

How does one disable this feature?

That agentic stuff is going to be a big deal. Probably the most interesting part of LLMs besides coding. Assuming it works well

I don't want any of this crap. We need to push for the right to opt out of AI features. All of this garbage should be opt in.

It is astonishing that the word “privacy” appears zero times in this announcement. There have been repeated controversies over exactly how Google sees just the URL I visit. Now they want to see the entire contents of multiple browser tabs?

Does anyone know if Chromium is spared of these features?

Can I block it as a site host? (Please don't respond about how I shouldn't want to and isn't it just like some other usecase that I'd obviously want not to block)

Taking a quick test spin. Seems to be enabled (on mac) by the existing "Gemini in Chrome" extension but requires a Chrome update. Which has a global shortcut (across all mac apps) of "Ctrl+g" (this can be changed). The additional AI features seem to come from the tab content's integration into the Gemini console (previously (I believe) the Gemini extension was merely a thin console to the Gemini service)).

The direct tab integration works by first showing the Gemini console (ctrl-g or Gemini icon in the mac system tray) where there is then a 'Share current page' icon below the text view.

This adds a blue border around the chrome window indicating the current page can be shared with chrome. It's not clear, but I believe the page content is only shared once a prompt is made intending to use the page's content. However, the share-enablement remains enabled for all tabs (all tab windows will have the blue border) until turned off in the Gemini consol. Again, it's not clear if just merely browsing these tabs will automatically share that content with Google.

The Gemini integration does not perform actions (can't ask it to do stuff directly with the site's content (navigate, click buttons, etc).

But direct summarization works well (try it on an HN comment page or news article).

Overall, I like this feature as long as I understand what and when things are being shared and ability to turn off easily.

This seems huge to me. As in: the initial release of Google Chrome huge. I don’t know if they can really pull off things they showed but if they do I’m sure this will be a massive success. Which is pretty sad considering the privacy implications for this. Imagine how much more data they will have on everyone. Scary.

Anthropic's "Claude for Chrome" pilot from a few weeks back spent most of the announcement article talking about security and threats from prompt injection: https://www.anthropic.com/news/claude-for-chrome

This announcement from Chrome themselves hardly mentions that at all.

I've just hard-pinned chrome v138 so ublock origin keeps working, so happy to hear it is also saving me from the AI features.

I suggest the people working on those features watch Neil Postman's "The Surrender of Culture to Technology": https://www.youtube.com/watch?v=hlrv7DIHllE

Part of me is actually excited for the acceleration of the end of the web, I even wrote https://punkx.org/jackdoe/zero.txt few days ago in frustration.

The sooner it ends the better.

I really don't want to see "summarization" tokens of other people's tokens.

Agentic browsing in Chrome is a really big deal. For the first time in 30 years the "user agent" will really act like one instead of being a mere browser.

In 10 years it will be passé to use the web in real time vs. having an agent acting on your behalf and distilling the information you want.

I've used probably 15 or 20 web browsers in my lifetime and all of them had the same barely searchable table of URLs as their only history view. Why couldn't we have full text search of the pages, or a view that reflects tab histories as some kind of graph, or UIs that support any kind of sorting? Instead it's 2025 and the solution is to attach an LLM slot machine to the front and drive engagement.

I'd be very open to any Firefox extension suggestions (or standalone applications that can consume a Firefox history) that makes it more searchable. I don't often need to search my browser history, but when I do the answer is rarely easy to find.

All of the other features look like a high potential for abuse, but with lots of glitz to make it seem essential to laymen.

if anyone wants a lightweight alternative I built an AI chrome extension (<1MB) that helps you fix grammar and improve writing wherever you write, draft emails on Gmail, reply to messages on LinkedIn etc.

All SOTA models are supported and you can use your own API key :)

https://jetwriter.ai

Does Google provide a version without AI?

For people who don’t want AI in their browser.

Calling it now, these features wont work with an adblocker enabled within 6 months.

Not that I asked for any of this anyway.

I cannot stress how much I don't want this.

The compromised password thing sounds quite handy for me. According to Google's password manager I have 139 compromised passwords and 357 reused passwords. They are mostly junk sites like commenting on the Croydon Advertiser or some such but even so being able to update them with one click would be nice.

Seems reasonable TBH.

I don't know the future of browsers given the trends in AI, but it seems fine to add an opt in ability to browsers to allow an LLM to access the current (or a set) of tabs. If it works it would reduce the amount of copy-paste, which seems like a good thing.

It's hardly a killer feature. I'm still going to use chatgpt (and gemini) a tremendous amount.

When the webpage AI argues with the browser AI, which argues with the OS AI which argues with the on-chip mainboard, CPU and GPU AIs, while the monitor AIs frantically try to make notes and the smarthome AI watches all of it and can only shake its metaphorical head.

My favorite Gemini feature is pasting an url with cultural events and say: add these events to my calendar.

Also works with images

They're all wearing shades of green.

Well, this sounds terrible. Asking the AI "what was the site where i saw a walnut desk" would mean that enough data has to be stored (locally but how long until there is a pro version where it stored and processed centrally). Isn't that data storage a security nightmare?

Last time I checked the context window for the embedded Gemini Nano was 1024 tokens. I hope they have reconsidered that limitation.

I would sooner stick pins in my eyes than trust Google with unfettered read access to my browser tabs.

There's no better time to switch to Firefox: http://getfirefox.com/.

Until FF pulls a dick move, at which point we sell all our possessions and go live in the mountains.

Why as user I should get excited for ai features? Why are they selling ai so much, show me what it solves ? Nothing

And yet even on the page about Chrome's new AI assistant is a cookie pop-up ...

> Gemini in Chrome can now work across multiple tabs, so you can quickly compare and summarize information across multiple websites to find what you need.

I noticed the other day that my own Chrome browser had begun doing something almost-but-not-quite like this, specifically in the context of browser-chrome-level search autocomplete.

Typing in the omnibar now seems to offer completion suggestions based on (seemingly) 1. feeding a fulltext extract of all my open tabs into a RAG; 2. doing a search of that RAG using the completion prefix as the query; and then 3. feeding the matched text-chunks to a local (Gemini Nano?) auto-completion suggestion model.

So if I have a tab open somewhere talking about e.g. what type of plants would grow well on a north-facing balcony in Vancouver, and then open a new tab and type "trellace " into the omnibox — then it'll suggest completions seemingly coming specifically from "ideas" it got by reading that open tab. Close the other tab, and those suggestions disappear.

(This might just be an experiment that was pushed to me, though I couldn't find anything specifically referencing it in chrome://flags.)

Very interested to see how well the agentic features work compared to ChatGPT’s cloud version. At the very least, I imagine bot detection/prevention (I.e., CAPTCHA) will be less of issue with Google’s strategy since the browser’s fingerprint will differ Chrome user to Chrome user.

Funny how this announcement comes days after Google learned that it didn't need to sell Chrome

Honestl, agentic browsing on grocery sites would be pretty great... But mostly because grocery sites have some of the worst UX in existence.

Oh, look a wild Microsoft Recall appears, sure I want the AI to know my browser history and what I do on the web

That reminds me to donate to Ladybird.

A lot of these new AI features can be helpful as long as the users' control the data. Is there an alternate world where Google might become the good guy here? In this world, as might be expected, the company making the worlds best spyware, wants to expand its spying.

Not a comment re: AI in Chrome. I did click into the video to hopefully get an understanding of the features,. but sadly did not. Sorta got an idea, but the jump cuts, super-quick overviews, trying to identify everyone speaking, etc. E.g. ...something about tabs... ...used to take 20 minutes, now seconds... ...working with OTHER Google products (WTH)... Hey Google, make a simpler, more accessible video that is "just the facts mam". Dozens of jump cuts in a few minutes is disconcerting, and may prove dangerous to some.

>>> And when it comes to a compromised password, Chrome warns me that my password got hacked and offers to change it for me. All I have to do is click once and that's it. I'm done

Do you folks hear yourself ? You want to change my password without me involving it other than clicking ?

>Combat more sophisticated scams with Gemini Nano

Fantastic, Googles AI will be fighting to stop the scams Googles advertising promotes to me...