OpenAI’s new Atlas browser has certainly got people talking.

Some are excited — calling it a “Google killer” and a glimpse of how we’ll all navigate the web in future. Others are alarmed — pointing to privacy concerns, data collection prompts, and the idea of handing over browsing history and passwords to an AI company.

Jason Grant described his experience as “a giant dark pattern.” Matthew Dunn was more balanced — impressed by the features, but quick to warn businesses off using it. He’s right: if you wouldn’t paste confidential data into ChatGPT, you probably shouldn’t browse the company intranet through Atlas either.

Search vs. synthesis

When people say Atlas will replace Google, they’re missing the point. It’s not search in the traditional sense.

A search engine indexes existing content and returns links that might answer your question. Atlas — and systems like it — go a step further. They synthesise an answer, combining what’s on the web with what’s in your conversation and what they’ve “seen” before.

As Data Science Dojo explains, search engines are designed to find information that already exists, while synthesis engines are designed to create new information.

Or, as Vincent Hunt neatly puts it: “Search gives you links. Synthesis gives you insight.”

That shift sounds subtle, but it changes everything: how we ask questions, how we evaluate truth, and how much we trust the output.

As I said in my recent talk on AI Transformation at the Bletchley AI User Group, “Generative AI is not a search engine. It doesn’t retrieve facts. It generates language based on probabilities.” Google doesn’t know the truth either — it just gives you the most common answer to your question — but AI goes a step further. It merges, rewrites and repackages information. That can be powerful, but it’s also risky. It’s why I believe the AI-generated results that many search engines now return as default are inferior to traditional results, based on actual information sources.

Without strong governance, AI may be repurposing outdated content or drawing on biased data. Transparency matters — because trust is the real currency of AI adoption.

Why Atlas matters

In OpenAI’s announcement, Atlas is described as “bringing ChatGPT anywhere across the web — helping you in the window right where you are.”

It’s not just a search bar. It can summarise pages, compare options, fill out forms, or even complete tasks within websites. That’s a very different paradigm — one where the browser becomes a workspace, and the assistant becomes a collaborator.

A step towards agentic AI?

So, is Atlas really agentic? In part, yes.

Agentic AI describes systems that can act rather than just answer. They plan, execute and adapt — working on your behalf, not just waiting for your next prompt.

OpenAI’s own notes mention an agent mode that can “help you book reservations or edit documents you’re working on,” as reported by The Verge.

Others, like Practical Ecommerce, describe Atlas as “a push into agentic browsing — where the browser is now an AI agent too.”

It’s not full autonomy yet — more like assisted agency — but it’s a clear step in that direction.

Why it still needs caution

As exciting as it sounds, Atlas isn’t designed for enterprise use. It raises valid concerns about data privacy, security, and trust. You wouldn’t give a work browser access to sensitive credentials, and the same logic applies here.

As Matthew Dunn notes, ChatGPT “produces better output than Copilot, but with less security and privacy.” That’s a fair trade-off for some users, but not for organisations handling confidential information.

So, by all means, explore it — but do so with your eyes open.

And yes, I’ll still give it a try I decided not to install it after all

For all the justified concerns about privacy and data handling, I’ll still give Atlas a try. Even though I have Copilot at work, I pay for ChatGPT Pro for activities that are not directly related to my confidential work.

Atlas might extend that usefulness into how I browse, not just how I prompt. The key, as ever, is knowing what data you’re sharing — and making that a conscious choice, not an accidental one.

[Updated 24/10/2025: After writing and publishing this post, I decided not to install Atlas. There are a lot of security concerns about the way the browser stores local data, which may easily be exploited. Nevertheless, both OpenAI Atlas and Perplexity Comet are interesting developments, and the narrative about the differences between an AI search (synthesis) and a traditional search is still valid.]

Featured image: created by ChatGPT.