Google is bringing agentic AI directly into Chrome. The world’s most popular browser just became an assistant. It can navigate websites. Fill forms. Complete multi-step tasks like booking travel or shopping on your behalf.
The new “auto browse” feature runs on Gemini 3. It’s available now to Google AI Pro and Ultra subscribers in the United States. Google is racing to fend off competition from AI-native browsers. OpenAI, Anthropic, and others are building them.
Traditional browser automation relies on APIs. Chrome’s agentic browsing is different. Gemini controls the browser like a human user would.
The AI can search for pet-friendly apartments across rental sites. It can plan trips by comparing multiple travel platforms. It can manage shopping carts. It pauses before sensitive actions. Completing purchases. Making posts. Things like that.
A persistent side panel maintains context across tabs. The AI integrates with Gmail, Calendar, YouTube, and Maps. It accesses your broader Google ecosystem.
Google is planning a “Personal Intelligence” memory layer. Chrome will remember past conversations. It’ll tailor responses accordingly. The feature is opt-in. That addresses growing concerns about how much personal data AI systems should store and access.
The agentic browser landscape is getting crowded. OpenAI’s Atlas. Perplexity’s Comet. Anthropic’s Claude for Chrome. Opera Neon. Norton NEO. BrowserOS. They’re all vying for market share. Each emphasizes different strengths. Privacy. Speed. Conversational search.
Google’s advantage? Distribution and ecosystem lock-in. Chrome is already installed on billions of devices. It’s deeply integrated with services users rely on daily.
But agentic browsing introduces new security vulnerabilities. Google highlights indirect prompt injection as the main threat. Malicious pages can hide instructions. They trick the agent into leaking data or initiating transactions.
The company says Gemini can’t directly access passwords. It can’t download files. It can’t run code. It can’t navigate to unrelated sites. A “User Alignment Critic” model reviews actions before execution. Sensitive-site checks run deterministically. Users must confirm all financial operations.
Google has layered defenses. It offers bounties up to $20,000 for bypassing safeguards. The company still concedes the system isn’t bulletproof.
Google is simultaneously promoting the Universal Commerce Protocol. It’s working with major retailers. Shopify. Etsy. Wayfair. Target. The goal is smoother AI-driven transactions.
Auto browse remains U.S.-only for now. It’s paywalled behind subscription tiers. It’s labeled as a preview feature. It’s rolling out gradually across desktop and mobile.
Agentic browsing could fundamentally reshape e-commerce. The paradigm shifts from users browsing stores to AI agents negotiating purchases on their behalf. That would upend online advertising. Affiliate marketing models. How merchants compete for visibility.
The rollout signals something bigger. AI agents are moving from experimental tools into core web infrastructure. That raises urgent questions about user trust. Data governance. Whether convenience will outweigh the new attack surface these systems create.
