AI Browsers are here - what are the pros and cons of Comet & Atlas?

Chat GPT Atlas & Perplexity Comet browsers have landed. Both look instantly familiar as they are built on the Google chromium platform and, beyond search, both can perform handy agentic workflow tasks (if you use them to log-in to your social platforms and email accounts: more on that later...) such as locating Linkedin profiles and auto drafting messages, finding discount codes and applying them to purchases, finding the specific part of a video with a talking point, and even generating strategy documents. Comet also presents itself as a "study buddy & tutor".

Anyone who follows my (somewhat sporadic) social posts will know that;

1. I'm an AI skeptic. I think there's many uses for AI at a basic level, but the public becomes desensitised to AI creative very quickly, the error rate in output remains too high, the tech presently presents unacceptable risks to data privacy, and IP infringements need to be addressed.

2. I'm a big fan of Perplexity. I've been using it for years and it's no surprise to me that Comet is streets ahead of Atlas.

Streets ahead, how?

Atlas sees agentic function as a feature to activate. Comet sees AI agents as a wholly integrated part of the browser. This means that, on Comet, agents act seamlessly. From most independent reports, the Comet agents also tend to be more effective across all functions; they get it right when locating emails or finding the right booking, and Comet also does this more quickly in most cases. Search answers have clear citations with links to sources - something Perplexity got right from the very start.

This is a real wake up call for Chat GPT as their valuation and broader perception as the defacto leader of the AI agents takes a real hit in this first round of AI powered browser wars. You can see the difference if you check out the side by side comparison video from Riverside's Stephen Robles below.

Atlas vs Comet

Proceed with caution

AI browsers need a full view of your online world to work well, which means they often see absolutely everything: this includes your browser history, files, and even sensitive work documents are all potentially accessible, making accidental leaks a real possibility.

You may even end up unintentionally sharing private content as AI’s all access pass means confidential messages, financial drafts, or personal data get swept up and sent to places you didn’t intend.

And then there are third-party data transfers: Depending on how your AI browser is built, your personal info might be processed in the cloud, where it’s exposed to external servers far beyond your local machine.

The main downside to having AI browsers perform tasks on your computer are the significant privacy and security risks. There is huge potential for data breaches, and there are vulnerabilities to new types of attacks like "prompt injections". These browsers break down traditional web security barriers to function, which can give malicious actors access to personal and confidential information.

AI-powered browsers give you a bunch of convenient and time saving auto-complete features, but before you let them handle your online tasks, it’s worth understanding the serious risks that come with the deal. The downsides — especially in terms of privacy, security, and reliability — are still hard to ignore.

Granting an AI browser access do your bidding can open the door to some nasty surprises:

Prompt injection vulnerabilities: Hackers have ways to sneak malicious instructions into images or hidden scripts. AI browsers can be tricked into downloading malware or, even worse, making scam purchases without your knowledge.

Broken security boundaries: To perform tasks across websites, these browsers often bypass the barriers meant to keep different web services isolated. It’s effective for time-saving, but it also makes it easier for cybercriminals to steal data or hijack accounts.

If you want to auto-complete tasks, you are required to cede control of your computer. The moment your AI browser goes autonomous, stopping it isn’t always straightforward. If things go wrong, it could end up accessing sensitive private info or landing on dodgy sites before you’re able to intervene.

There is a reason that Google & Microsoft are yet to roll out these agentic features on their browsers - I imagine it is that they are playing the "second mouse" card. Let Open AI and Perplexity take the legal heat for early shortcomings, watch and learn before they proceed to market and curb some of the risks of costly litigation over a system which technology companies (and markets) have bet the farm on.