“It’s nearly impossible these days to attend a technical conference or standards meeting without the conversation turning to AI.”

In identity and cybersecurity circles, especially, the questions are overwhelming in number. How should an AI agent be identified? How do we manage delegation to an agent, and to whatever other systems that agent might enlist to complete its task? How do we limit the scope of those actions so that “acting on my behalf” doesn’t quietly turn into “acting far beyond what I intended”?

These are big questions that we don’t yet have standard answers to, and they show up in many places across the technology stack. That means a) you can approach them from a variety of angles, and b) people working on those angles may not always know what’s happening in the others. Still, you have to start somewhere, and right now I’m looking at this from the perspective of the web.

You can Subscribe and Listen to the Podcast on Apple Podcasts, or wherever you listen to Podcasts.

And be sure to leave me a Rating and Review!

The World Wide Web

The web platform has strong architectural principles (see the Ethical Web Principles and the Web Platform Design Principles, for example) about who acts, who decides, and who ultimately controls what happens online. So one of the questions I’ve been pondering is how the recent wave of “AI browsers” fits into those principles.

For the purposes of this discussion, an AI browser is simply a web browser that incorporates AI-supported features. Those features might include summarizing the contents of a page, synthesizing information across multiple sites, or providing a chatbot-style interface that allows users to request actions in natural language.

Sometimes those actions are fairly small: summarizing a document, finding a quote on a page, or suggesting related material.

Increasingly, though, these features involve the browser doing things on the user’s behalf. Finding information. Filling out forms. Comparing products. Potentially even making transactions.

The more those capabilities expand, the more they start to stress an important concept in the web architecture: the definition of a Web User Agent.

Before digging into why that matters, it’s worth noting something about the “AI browser” label itself. Referring to a product or feature as “AI” is usually temporary. It lasts as long as marketing departments find it useful. Eventually, the feature stops being special and simply becomes part of the baseline product.

Spellcheck was once a major innovation. Now it’s just expected.

The same thing will happen here. Today, we talk about “AI browsers.” In a few years, we will probably just call them browsers.

Which is exactly why this is a good moment to ask what their behavior means for the web’s underlying principles.

The Browser’s Traditional Role

The web architecture has always placed the browser in a very particular role.

In standards language, the browser is a user agent—software that retrieves web content and presents it to the user while acting on the user’s behalf.

The key phrase there is on the user’s behalf, but historically that phrase has had a fairly narrow interpretation. A browser mediates interaction between a person and the web. The user clicks a link, and the browser retrieves a page. The user fills out a form, and the browser submits it.

The browser acts as an intermediary, not as a decision-maker.

The W3C’s Priority of Constituencies reinforces this model. In that framework, the needs of the user come first. Then come the needs of web page authors, then user agent implementors, then specification writers.

The ordering matters because it reflects the underlying philosophy of the web:

The web exists for its users.

Browsers enforce that principle in many ways. They manage permissions for sensitive capabilities like camera access or geolocation. They work to protect users from malicious content. And of course, they provide indicators that help users understand where they are on the web and what they are interacting with.

tl;dr – the browser is expected to be designed to represent the user’s interests.

Where AI Browsers Start to Stretch the Model

AI features don’t automatically change that role.

If a browser summarizes a page you are reading or answers a question about its content, it is still behaving like a traditional user agent. It is helping you interpret what you have already chosen to view.

But things become more complicated when the browser begins to take actions that extend beyond the user’s immediate interaction.

Consider a few examples that are already appearing in prototype systems:

• Asking the browser to research a topic across dozens of sites
• Requesting the browser to monitor price changes and notify you later
• Instructing the browser to find and book travel within certain parameters
• Allowing the browser to navigate services and fill out forms autonomously

In these scenarios, the browser is no longer simply mediating interaction. It is executing delegated intent. The user expresses a goal once, and the browser may perform many actions afterward to achieve that goal. These are words that get identity people both excited and anxious, as we’re not yet confident in how exactly to achieve that delegated intent in a safe and standardized manner. (If you’re particularly interested in intent, you might be interested in this interview with Eve Maler on the Identerati Office Hours.)

This changes the nature of the browser’s role in subtle but important ways.

Delegation Introduces More Questions

So let’s pause for a bit on that exciting and anxiety-inducing concept. Delegation is not a new concept on the web. We already allow applications to perform actions on our behalf through APIs, tokens, and service integrations.

What is new is the idea that the browser itself may become a delegated actor.

That raises several architectural questions.

What counts as user intent?

Web interactions traditionally reflect direct user input: clicks, navigation, or form submissions.

Delegated browsing introduces a new pattern where intent is expressed once and actions unfold later. Determining how far that original intent extends becomes a design question.

How does consent work?

Permission systems on the web are largely built around real-time user decisions. A site asks for access to the camera or location, and the user approves or denies the request.

If a browser is acting autonomously hours after the user issued a request, how should that consent be interpreted?

Who is accountable for actions?

When a browser submits a form or interacts with a service automatically, the action is still attributed to the user. But the user may not be aware of every intermediate step taken to achieve the final result. I’m sure the browser’s terms of service will still assign responsibility to the user, but I think we’ll see some heated debates on that going forward.

Ultimately, maintaining transparency and accountability becomes harder as the browser takes on more initiative.

The Blurry Line Between Browsers and Bots

Another interesting effect of AI-driven browsing is that it blurs a long-standing distinction on the web.

Historically, the web has treated browsers and bots very differently.

Browsers represent human users interacting with sites. Bots, such as crawlers or automated scripts, interact with the web at scale and are often subject to restrictions or special handling.

AI browsers occupy an ambiguous space between those categories.

They are controlled by a human user and operate through a browser environment. But they may also perform automated actions that resemble bot behavior, such as navigating many pages, collecting information, or interacting with services programmatically.

As these capabilities expand, websites and standards bodies may need to reconsider how they draw that distinction.

Why This Conversation Matters Now

None of these questions has an immediate, clear answer. Welcome to living in interesting times.

But the web has a long tradition of addressing architectural challenges early, before behaviors become too deeply embedded in the ecosystem. That is why the rise of AI browsers is worth examining now, while these capabilities are still emerging.

The core issue is not whether browsers will incorporate AI features. That outcome is almost certain. Over time, these features will stop being marketed as “AI” and will simply become normal browser functionality.

The real question is whether the browser will remain primarily a mediator between users and the web, or whether it will evolve into something closer to a proxy acting within the web on the user’s behalf.

If the latter becomes the dominant model, the definition of a web user agent—and the principles that guide its behavior—may need to evolve as well.

📩 If you’d like to be notified of new posts rather than hoping you catch it on social media, I have an option for you! Subscribe to get a notification when new posts go live. No spam, just announcements of new posts. [Subscribe here] 

Transcript

Introduction

00:00:00

Welcome to the Digital Identity Digest, the audio companion to the Spherical Cow Consulting blog. Each week explores key developments in digital identity—from credentials and standards to browser behavior and policy shifts.

If you work in digital identity but don’t have time to track every emerging trend, this is your shortcut to clarity.

00:00:26

Let’s get into it.

---

The Growing Presence of AI in Identity Conversations

00:00:31

AI is now a constant topic at technical conferences and standards discussions. In particular, identity and cybersecurity communities are asking urgent questions, such as:

• How should an AI agent be identified?
• How do we manage delegation to that agent?
• How do we limit its scope of action?

00:01:06

These questions remain largely unanswered. Moreover, they appear across multiple domains:

• Infrastructure
• Identity systems
• Application security
• Governance and policy

As a result, efforts are often fragmented and poorly coordinated.

---

Why Start with the Web?

00:01:39

Every complex problem needs a starting point. Here, the focus is on the web platform.

The web has long-established architectural principles defining:

• Who acts
• Who decides
• Who controls outcomes

00:01:56

This raises an important question:

How do AI-powered browsers fit into these existing principles?

---

What Is an AI Browser?

00:02:05

For this discussion, an AI browser is simply a browser that includes AI-supported features.

These features may include:

• Summarizing web pages
• Synthesizing information across sites
• Chat-style interfaces for natural language requests

00:02:37

Increasingly, however, these browsers go beyond assistance and begin to act on behalf of the user.

Examples include:

• Filling out forms
• Comparing products
• Making transactions

This shift begins to challenge a core concept: the definition of a web user agent.

---

A Quick Note on “AI” as a Label

00:03:08

The term “AI” is often temporary. Historically:

• Spellcheck was once considered innovative
• Today, it’s simply expected

00:03:24

The same will likely happen with AI in browsers. Eventually, “AI browsers” will just be called “browsers.”

This makes now the right time to examine the deeper implications.

---

The Traditional Role of the Browser

00:03:46

In standards language, a browser is a user agent—software that retrieves and presents web content on behalf of the user.

Traditionally, this role is limited and clear:

• The user clicks → the browser retrieves
• The user submits → the browser sends
• The user navigates → the browser requests

00:04:23

In short, the browser acts as an intermediary—not a decision-maker.

---

The Web’s Core Design Principle

00:04:31

A foundational concept in web architecture is the priority of constituencies:

1. Users
2. Web page authors
3. Browser developers
4. Specification writers

00:04:58

This hierarchy reinforces a central idea:

The web exists for its users.

Browsers support this through:

• Permission controls (camera, location)
• Security isolation between sites
• Clear indicators of context and identity

---

Where AI Begins to Change the Model

00:05:44

Not all AI features disrupt this model.

For example:

• Summarizing a page
• Answering questions about visible content

These still support user-driven interaction.

00:06:01

However, things shift when browsers begin acting beyond immediate input.

---

From Interaction to Delegation

00:06:09

Emerging capabilities include:

• Researching across multiple sites
• Monitoring prices over time
• Booking travel automatically
• Completing workflows independently

00:06:36

Here’s the key difference:

• Traditional browsing = interaction
• AI browsing = delegation

The user sets a goal once, and the browser executes multiple steps afterward.

---

New Architectural Questions

00:06:53

This evolution introduces several critical challenges.

What Counts as User Intent?

00:07:09

Previously, intent was immediate:

• Clicks
• Form submissions
• Navigation

Now, intent may be:

• Expressed once
• Executed over time

This raises questions about scope and limits.

---

How Does Consent Work?

00:07:34

Current permission systems rely on real-time decisions.

But with AI:

• Actions may occur hours later
• Users may not see each step

This complicates how consent is defined and enforced.

---

Who Is Accountable?

00:07:50

Even if the browser acts autonomously:

• Actions are still attributed to the user
• Visibility into intermediate steps is limited

Maintaining transparency becomes significantly harder.

---

Blurring the Line Between Browsers and Bots

00:08:22

Traditionally:

• Browsers = human-driven
• Bots = automated systems

00:08:45

AI browsers sit in between:

• Controlled by humans
• Capable of automated behavior

They may:

• Navigate multiple pages
• Collect data at scale
• Interact programmatically

00:09:04

This creates a gray area that challenges existing distinctions.

---

Why This Matters Now

00:09:18

The web has a strong tradition of addressing challenges early—before they become entrenched.

AI browsers are still emerging, making this the ideal time to ask:

• What should a browser be?
• What responsibilities should it hold?

---

The Bigger Question

00:09:48

Looking ahead, one question stands out:

Will browsers remain intermediaries—or evolve into autonomous agents?

If they become agents:

• The definition of a user agent may need to change
• Core web principles may need to evolve

---

Closing Thoughts

00:10:18

Browsers are central to the web ecosystem. Any shift in their role has wide-reaching implications.

This is a conversation worth having now—before new behaviors become the default.

00:10:26

Stay tuned. This topic is far from settled.

---

Outro

00:10:38

That’s it for this week’s Digital Identity Digest.

If this helped clarify things—or sparked new questions:

• Share it with a colleague
• Connect on LinkedIn
• Subscribe and leave a review

You can also find the full written post at sphericalcowconsulting.com.

Stay curious. Stay engaged. Let’s keep the conversation going.