Google's 'Project Jarvis' AI leak hints at a new web assistant for Chrome

When Techloy first reported on Google's expected “computer-using agent,” Project Jarvis AI, we anticipated it would roll out gradually, coming to Gemini Advanced ~1 million subscribers first.

But in a classic Google twist, the tech giant “accidentally” leaked the tool ahead of time on the Chrome Web Store, only to swiftly pull it down.

For the lucky few who managed to download the extension, excitement quickly waned, says Androidpolice, as Jarvis required specific permissions that kept it from working, leaving early adopters with only a glimpse of its promise.

Jarvis, likely powered by Google’s upcoming Gemini 2.0 AI model, is expected to tackle tasks like booking flights, researching products, and finding shopping deals directly in Chrome.

This autonomous agent is rumoured to not just respond to voice commands like typical assistants; it will help navigate pages, click buttons, and fill in fields by analyzing screenshots of the current webpage—bringing “hands-free” browsing to a new level.

This accidental release aligns with recent moves by other tech giants exploring computer-controlling AI. Anthropic’s Claude, Apple’s “onscreen awareness,” thanks to Apple Intelligence, and Microsoft’s Recall also perform browser-based tasks, yet Jarvis seems poised to offer a broader suite of web-based capabilities.

While Apple’s and Microsoft’s tools focus on screen awareness and screenshotting, respectively, Jarvis aims to automate complex, multi-step web interactions, potentially setting it apart from competitors.

With “Project Jarvis” as its codename, there’s a possibility Google may rename it before launch. Privacy advocates may be on alert as the project moves forward—especially with Microsoft’s Recall already facing criticism for storing sensitive screenshots. Jarvis could face similar scrutiny, but for now, all eyes are on its potential December debut, as users anticipate whether it will redefine how AI integrates into daily web use.