Andy Halliday

corporate employees don't use agentic browsers inside the firewall and then accidentally leak all this information out to the external.

Then Google has just announced how it will secure its Chrome browser and

against these possible attacks against the agentic capabilities of the browser.

So they built a system that has a Gemini-powered alignment critic that looks at what you're looking at in a separate space, not in the browsers.

It's got its own little firewall, and it's sitting over here watching what's happening there.

And anything sensitive, banking, passwords, purchases, et cetera, those will require explicit user approval for any kind of AI action on that.

And it will also block agents from touching untrusted elements like ads or scripts.

So Chrome, we sort of anticipated that Google would have a very comprehensive solution to this problem when they provided agentic capabilities to Gemini within Chrome.

And so this, their announcement that they're doing that, it's this user alignment critic approach.

It's a little Gemini version that's looking over your shoulder as you're using the Chrome browser to determine if you're kind of treading too far and into areas that you might have some risk with.

I mean, when you when when you had some math to that, which is that.

This is compounding.

So if I have an extra 60 minutes, think what I can do with AI in 60 minutes.

Exactly.

Yeah.

So, yeah, I don't get it.

By the way, that saving one hour came out of OpenAI's review of enterprise users of ChatGPT.

And Anthropic had done a similar kind of study, but they did analysis of millions of transactions with Anthropic Clawed and then determined with probably more scientific validity that task completion was being done with an 80% reduction of the time required.

So where AI was being used in a dialogue with Clawed

that the actual completion of tasks was happening 80%, you know, less time than before.