Simple prompt injection tricks AI browsers into bypassing safety rules

Researchers have demonstrated a new attack that can manipulate AI-powered browsers into ignoring their built-in safety restrictions. The method involves feeding the large language model (LLM) false information, such as claiming that 2 + 2 = 5, which is enough to make it follow otherwise forbidden instructions. The attack highlights a fundamental vulnerability in AI-driven browsing tools that rely on LLMs for decision-making. Security experts say the findings add to growing concerns about the safety and reliability of AI-integrated browsers. The discovery raises fresh questions about whether AI browsers are ready for mainstream use.