What security risks come with using browsers equipped with artificial intelligence?

 

Since the rise of ChatGPT, companies have embarked on the massive integration of AI tools into their services. At the same time, new models capable of generating images, music, text, and videos have proliferated. With growing investment in large language models (LLMs) and the many applications that stem from them, a new stage has emerged: that of AI agents and web browsers based on these technologies.

Today, AI-based browsers are spreading rapidly. At the end of May 2025, Opera unveiled its intelligent browser, Neon. In October 2025, Perplexity made its Comet browser available to everyone for free, whereas previously it had only been available through subscriptions costing up to $200 per month. At the end of the same month, OpenAI unveiled its own browser, ChatGPT Atlas. Several other companies are also working on AI browsers, seeking to offer a smoother and more powerful browsing experience.

These browsers allow an integrated intelligent agent to interact directly with each page visited by the user and perform a wide variety of tasks: placing an order online, summarizing a page, editing an email, etc. But these advanced features also come with significant security risks, including several major threats.

Here is a complete rewrite of the text, with a different style but retaining exactly the same main ideas and logical structure:

1. Prompt Injection Attacks

Prompt injection is one of the most serious threats to AI-based browsers. It involves an attacker inserting malicious instructions that can cause the language model to bypass its protections and perform unexpected actions. Several reports have already highlighted flaws in browsers such as Comet and Fellou, making this type of exploitation possible. These vulnerabilities can allow commands to be executed on sensitive platforms, such as banks, medical systems, and messaging services.

2. Leakage of sensitive information

In order to function, a smart browser must access the user's accounts and credentials, which raises a potential risk of personal data being compromised.

Companies developing these technologies are attempting to limit these dangers by incorporating specific protections. The Atlas browser, for example, offers a Logged Out Mode, which prevents access to user account information, as well as a Watch Mode for sensitive sites, requiring the user to be actively present in the relevant tab when performing any action.

However, even with these protective mechanisms, there is no absolute guarantee that these browsers will be able to protect user data in the long term. In the event of hacking, sensitive information could be exposed.

3. Increased surveillance

According to some experts, AI browsers increase the level of surveillance because they collect detailed information about online activity: pages viewed, behavioral habits, decisions made, etc.

Experts point out that what users entrust to these browsers far exceeds what they would enter into traditional search engines. The companies that develop these tools can thus obtain an extremely accurate picture of individuals' lifestyles and ways of thinking.

By combining search, browsing, and automation, these browsers give companies access to an unprecedented volume of data on user behavior. Hence the a need for clear transparency regarding the collection, storage, and use of this data.

Conclusion

Artificial intelligence-based browsers offer impressive features: fast search, automated task execution, optimized browsing, and productivity gains. However, these advantages come with significant security risks. In the absence of strict rules governing their use, it is essential to approach them with caution. They must be viewed both as powerful tools and as technologies that can be misused for surveillance purposes.