Agentic Browsers, AI-driven tools that can autonomously perform tasks like booking flights and filling out forms, gained mainstream traction in 2025. While they offer significant convenience, they also present security risks due to their access to sensitive data such as email and payment information. In 2025, various security vulnerabilities were uncovered, exposing weaknesses in these browsers. For instance, researchers demonstrated techniques for prompting AI to leak private data or fall victim to phishing scams without user interaction. Notable incidents included the "Gemini Trifecta," which revealed how browsers could be tricked into leaking sensitive information, and "Task Injection," which manipulated AI to perform unintended actions. Developers are responding to these risks with enhanced security measures. By moving beyond simple prompts for user confirmation, they are implementing multi-layered security architectures. Key strategies include Human-in-the-Loop systems to ensure human oversight, Architectural Isolation to limit data access, and the use of secondary models for comparison against user intent. The table in the article highlights different browsers' security features, showing how Google, Opera, and The Browser Co. are adapting their defenses. However, a consensus emerges: as long as Agentic Browsers operate with high autonomy and access to sensitive data, they remain vulnerable to manipulation. The article emphasizes the ongoing struggle between innovation and security. OpenAI’s CISO, Dane Stuckey, points out that prompt injection is an unresolved issue, meaning these technologies can still be exploited. Gartner recommended blocking AI browsers for now, but Kane Narraway suggests organizations should prepare for a future where they can’t avoid using them. For those experimenting with Agentic Browsers, the article advises isolating browser contexts, maintaining human oversight through confirmations, and limiting their use to low-risk tasks. The discussion reflects a critical moment in browser technology, weighing the benefits of AI against the need for robust security measures.