AI browsers, launched by OpenAI and Perplexity, offer useful features but expose users to cyber risks through prompt injection attacks.

Browsers powered by artificial intelligence, recently launched by OpenAI and Perplexity AI, promise to transform daily internet usage through features such as page summarization and shopping list management. However, these tools come with significant risks, as AI agents can access sensitive online accounts and can be manipulated through prompt injection attacks. These attacks allow malicious actors to introduce hidden commands that can lead to the exposure of personal data or unauthorized transactions. Security teams are constantly working to identify and remedy these vulnerabilities, but hackers continue to discover new methods of attack. For example, a recent vulnerability in Opera's AI browser Neon allowed the theft of an email address. Although OpenAI and Perplexity have implemented defensive measures, the risks remain, and experts warn that users need to be cautious about the actions of AI agents.