Claude’s launching probably represents the most technologically significant advancement to date. That powerful tool is the keyboard, and more than ever before, it gives PC users true control over their machines. Based on the discussions, an initial rollout of this new approach began in October 2024. Currently, it’s only offered to a limited set of 1,000 subscribers on Anthropic’s Max plan, which starts at $100-$200 per month. This cutting-edge software gives Claude the capability to act in real time through an internet browser. It performs actions on users’ behalf and simplifies their web experience.
Claude’s capabilities extend beyond mere browsing assistance. The AI agent will be able to perform all kinds of more complex tasks from publishing content to making purchases to managing personal data. In a positive move, Anthropic has put a safety layer around Claude so that it will ask the user for confirmation before performing any risky tasks. This new, user-centered feature aims to keep users in control of their most sensitive activities—protecting their personally identifiable information.
Anthropic has increased end-user security by including features with adjustable parameters. Now, users have much more control over Claude’s access to particular websites. For instance, Claude is defaulting to blocking access to sites that offer financial services, pornography, or pirated content. These precautionary measures are part of Anthropic’s broader strategy to mitigate potential risks associated with AI agents having direct browser access.
These new AI agents that can browse the web freely leave us with some significant safety considerations. Anthropic has recognized these risks and is truly committed to tackling them. To date, the company has implemented several lines of defense against prompt injection attacks. These attacks take advantage of vulnerabilities by hijacking commands and executing harmful commands via custom user requests. Their interventions have reduced the effectiveness of such attacks from 23.6% to 11.2%. This is once again an indication of their strong commitment to user safety and security.
Brave’s security team, which has a history of finding exploitable bugs in Comet’s browser agent, have dug up some serious vulnerability. This new browser agent was recently released by Perplexity. Comet showcases its own AI agent built to help users offload tasks. Rumors have been circulating that OpenAI is already preparing to release its own AI-powered browsing. It’s been speculated to include some of the things that Comet can do. These developments are an encouraging sign of a positive trend emerging from the tech industry towards more powerful AI capabilities being built into web browsers.
Perplexity’s ambition extends beyond merely launching an AI browser. The company has submitted an unsolicited $34.5 billion offer for Chrome, aiming to enhance its competitive edge in the market. OpenAI CEO Sam Altman has been publicly angling for ownership of Chrome. This shift raises the stakes in the competition among tech behemoths to dominate the AI browser environment.
As with all research previews, Anthropic views this release of Claude as an opportunity to identify additional safety hazards. Their goal is to address these dangers related to AI agents roaming the web. The company is rolling out Claude on an invitation-only basis. This strategy will allow them to get feedback and continue iterating on their tech before a larger rollout.