Battle for browsers: Anthropic vs Perplexity and OpenAI
It seems Anthropic wants to transform the concept of browser technologies. Claude for Chrome enters closed testing for 1,000 Max-plan subscribers.
This Chrome extension creates a sidebar with Claude who understands the context of everything happening in the browser. The agent doesn’t just observe. It performs actions on your behalf. Booking tickets, filling forms, managing email — all automated through natural dialogue.
Browsers are becoming the new battlefield of tech giants. Perplexity launched Comet with integrated AI agent. OpenAI is preparing its own browser with similar functions. Google is implementing Gemini integrations into Chrome.
Anthropic also conducted extensive testing on 123 test cases and 29 different attack scenarios. Without protective mechanisms, attack success rates reached 23.6%! Example of successful attack — malicious email made Claude delete all user’s mail without confirmation.
Anthropic says it has effective multi-layered protection. Site-level permissions allow controlling access to specific sites. Action confirmations require confirmation before high-risk actions — posting, purchases, personal data transfer. Even in autonomous mode, critical protective mechanisms remain.
System prompts have been rethought for maximum security. Claude receives detailed instructions for handling sensitive data before executing user commands. According to them, this isn’t just a feature, but an architectural solution built into the technology foundation.
Access cost — from $100 to $200 monthly for Max-plan — positions the product for professional segment. A waitlist for interested users is also open. “Automation through natural dialogue” sounds convenient. But can this make us less capable of independent actions? Perhaps the inefficiency of human actions has its own value.
