The New Browser Wars?

Hello fellow humans! Today, there is big news from the frontier AI firms launching browser tools. Browser wars 2.0 appears to be heating up. But the important stories are how chatbots are unethical therapists and some reasons why enterprise AI deployments are floundering.

Yesterday, OpenAI launched a new web browser called ChatGPT Atlas. This new browser integrates the chat interface into a sidebar, and expects you to interact with the browser in that sidebar, rather than the main browser viewer. I found it a bit disorienting interacting with the sidebar while content renders in the main browser window. But the main feature here is that you can instruct the browser to do work for you and it can autonomously navigate through a complex website for you to do it. Agentic work can be done either in logged in and a logged out mode. The logged in mode carries your information and can theoretically input your personal information and can begin the booking or purchase process, but will require your input and approval before payment.

OpenAI’s opening offer is surprisingly full-featured browser and really commits to support for agentic work. For example, it can do work for you in the browser such as filling out forms, clicking through multi-screen interfaces. It also leverages the context of your ChatGPT history, so it knows you to whatever degree you’ve revealed yourself to ChatGPT. It also provides an entry point for OpenAI into the advertising market to compete with Google, although this is more of a win for OpenAI than it is for you.

Leverages the context of your ChatGPT history in a browser, and considering the troubling way that people have been using ChatGPT for deeply personal subjects, this may be some cause for alarm. OpenAI will almost certainly be using those chat histories to target advertising to us. This also radically changes how we interact with the internet as the chatbot becomes an intermediary between us and the web content. Ultimately, this will break a lot of the internet as it is built right now, and companies like Expedia will will need to retool around customer outcomes rather than human-visual pages for navigation.

There is also a risk for prompt injections, and the chatbot acknowledges this; it will tell you when it thinks there may have been a prompt injection. It dutifully resets and restarts the work. But certainly, people will test this and find workarounds.

This is an impressive first foray into the browser market, and definitely sets up OpenAI as a competitor to Google for mindshare and advertising revenue, something that OpenAI desperately needs as more investors get nervous about an AI bubble bursting. People already deeply involved in their ChatGPT relationships may be ready to jump into this new paradigm of experiencing the internet, but for people with a more traditional mental model of the web may find it disorienting and unwelcome. But this is a bellwether experiment to find out which way the internet will evolve.

Very similar to everything I just covered in the ChatGPT Atlas, but it does not have the agentic element to it — it cannot do work for you — yet. If privacy is a concern for you, Google is already a known quantity, whereas OpenAI is still in a formative state, and can evolve in a range of different ways.

Because the Gemini integration doesn’t manipulate the browser content window, it may seem like it is doing less, but on a hypothetical flight booking search, Gemini returned results much faster than Atlas. And while it didn’t offer to take me straight to booking, it did provide links to booking pages.

Google’s deep relationship to internet content creators and app builders may offer it a distinctive edge over OpenAI here. Google has already been wrestling with the complex issues of internet content and delivering traffic for those content creators, so they may be better positioned than OpenAI to strike a win-win balance that keeps the entire internet ecosystem humming.

Overall, I find Gemini less intrusive and less inclined to take over the browsing experience, but we’re in the early stages of this evolution, so we’ll have to see where this experiment takes us.

We’ve been following the troubling trend that many people are using ChatGPT for free therapy, even though using chatbots this way actually makes people feel more lonely. So it is all the more alarming when a Brown University study found that AI chatbots “systematically violate ethical standards of practice established by organizations like the American Psychological Association.”

The study, led by Zainab Iftikhar, a Ph.D. candidate in computer science at Brown, evaluated a series of prompts to survey how LLMs respond to various therapy-related conversations. The goal is to help provide a framework to “determine whether such strategies could help models adhere to ethical principles for real-world deployment.”

Iftikhar and her team crafted a “practitioner-informed framework of 15 ethical risks to demonstrate how LLM counselors violate ethical standards in mental health practice by mapping the model’s behavior to specific ethical violations.”

Among the 15 violations, it found that the LLMs exercised:

The research was presented yesterday, October 22, 2025 at the AAAI/ACM Conference on Artificial Intelligence, Ethics and Society.

Yesterday’s segment on prioritization highlighted some important frameworks for thinking about strategy, and in my mind, it connected to the latest Lenny’s Podcast episode where Lenny interviewed Nicole Forsgren and they talked about how to measure productivity, and by extension, how to measure success.

As much as we talk about productivity and value, they are notoriously difficult to measure, because they mean different things to different parts of the organization. The CFO thinks of value very differently than the Marketing Director, or the CPO, or the CEO. That’s the nature of complex organizations.

Click to read the full article.