The Double-Edged Sword of Al Browsers

Convenience vs Security Risks

- 

Browsers were once simple portals to the web. Now, they think. They summarise, automate, and sometimes even act. AI-powered browsers promise to make information retrieval effortless, freeing users from the grind of manual search and tab overload.

But as these browsers evolve from passive tools to autonomous agents, the line between assistance and intrusion begins to blur. Convenience, it seems, comes at a cost - and it's often paid in privacy and security.

New Capabilities, New Threat Surface

Today's AI browsers, such as those built by Opera, Perplexity, and Anthropic, integrate 'agentic' features that let the browser act on behalf of the user. Whether it's filling out forms, summarising PDFs, or fetching data across multiple tabs, these tools redefine multitasking. However, each new function adds complexity - and risk.

Security researchers have been showing attacks based on prompt-injection in which malicious web content, with four hidden instructions, trick an Al into leaking information or otherwise perform hazardous acts. Early in 2025, there were reports of AI browsers executing code as part of a web page (for example, Opera Neon implements such an agentic prototype), which raised concerns over unaccounted automation.

Similarly, cybersecurity experts from Malwarebytes and BrightDefense showcased 'CometJacking' exploits - deceptive URLs that manipulate Al agents to share session data from other tabs. In other words, attackers no longer need to hack software; they just need to hack the language the software understands.

Privacy and Supply-Chain Concerns