For many years, Google Chrome has been the leading web browser across various devices, including laptops and mobile phones. However, with the rise of generative AI, a new type of web browser called “agent browsers” is emerging, designed to complete tasks with minimal user input. While this sounds handy, security experts are raising concerns about the significant privacy and security risks these browsers could pose.
Those who are concerned about their personal data may find these new browsers more of a liability than a convenience due to the potential risks involved.
What is a web browser?
To begin with, a web browser is essentially an application on your computer, tablet, or smartphone that opens up access to the Internet, enabling you to search for information via platforms like Google or visit various websites. There’s a fair chance you’re reading this right now in your browser. Popular options currently include Google Chrome, Apple Safari, Microsoft Edge, and Mozilla Firefox.
What is an agent web browser?
An agent web browser is a standard web browser enhanced with artificial intelligence, offering unique capabilities through generative AI. The term “agent” indicates that the browser’s AI can act autonomously on your behalf, managing tasks you set for it like a personal assistant.
Examples of what these agents can do include researching information, booking hotels on travel sites, filling out online forms, or making purchases on e-commerce platforms like Amazon.
This category of browser is brand new, and its complete range of features is still being developed. Unfortunately, while many developers are racing for dominance in the agent browser space, the serious privacy concerns these browsers bring could make them a burden for users who value their personal information.
Dangers of agent web browsers
Brave, a web browser focused on privacy, has reported vulnerabilities in Comet, a new agent browser created by Perplexity. Their findings suggest that hackers can utilize a method called “indirect prompt injection” to launch attacks, allowing malicious agents to access banking sites, retrieve saved passwords, and potentially steal sensitive information from a user’s device, all without their knowledge. Perplexity has responded with a plan to quickly address these issues.
Similar concerns have arisen regarding Atlas, another agent browser from OpenAI. Reports by antivirus developers point out that the prompt function in Atlas might be taken advantage of to circumvent security measures and give commands for harmful activities, thereby also exposing sensitive data. OpenAI has denied these claims, asserting that the AI behind Atlas cannot access files or applications.
Underlying issues with agent browsers
It’s not just a coincidence that both OpenAI and Perplexity’s browsers share similar security weaknesses. The fundamental design of these agent browsers seems to have inherent flaws.
This situation arises from the reality that agent browsers can carry out personal tasks and handle sensitive data without user oversight. Once you start using one of these browsers, they essentially gain access to your entire digital life. You can provide them with instructions, but there’s no assurance that these browsers will always follow them correctly.
For instance, even if the AI manages to buy an item, it may not be what you intended. You might unwittingly authorize it to transfer your funds to an unintended account. Managing your digital life can become tricky, leading to potential privacy violations, and often, you might only realize something is amiss when it’s too late.
Even in cases where no hacker is involved, there’s still no guarantee that the AI will function as intended. The one fact we do know about generative AI is that it can be misleading, misrepresenting what it does and even inventing details.
A word of caution
For the moment, it’s advisable to steer clear of agent web browsers. They’re simply not refined enough yet. Avoid downloading or signing in with your own accounts, and don’t share any personal information. It’s best to hold off until some of these security concerns are adequately addressed. The risks associated with using agent browsers far outweigh the minimal benefits they might offer.
An unavoidable trend
Regrettably, the rise of agent web browsers can’t be ignored. Right now, you’ll find options like Atlas from OpenAI or Comet from Perplexity. However, these kinds of browsers are likely to become standard practice soon.
Both Google and Microsoft are working on adding agent-like features to their respective browsers, Chrome and Edge. Meanwhile, Mozilla, the developer of Firefox, is also exploring AI capabilities, though they claim users have the option to disable these features. As for Apple, there haven’t been any announcements regarding agent upgrades for Safari, and considering the company’s current stance on AI, upgrades seem unlikely in the near term.
Agent web browsers present an intriguing yet risky novelty that has suddenly captured consumer interest. Unless developers can significantly enhance their security features before widespread adoption, hackers will likely find new and effective strategies for data theft. To safeguard your information, it might be best to either disable these features where possible or choose browsers that do not incorporate agent AI at all.
