What the heck is Agentic Identity?

Identity is a very overloaded term.

The fundamental problem with how we’re discussing agentic identity is that we’re using the word “identity” to describe very different things.

Web Bot Auth, OAuth, the newly announced Personal Agent Protocol (PAP), and Know Your Agent (KYA) are all part of the conversation. But they deliver different information and solve different problems.

For one company, identity means proving that a request came from a particular AI agent. For another, it means recognizing a returning user through a persistent identifier. For a third, it means knowing the verified legal identity of the actual person behind the agent.

All are forms of identity, but they provide very different levels of trust and liability.

And importantly, neither AI platforms nor businesses care about identity for identity’s sake.

AI platforms care about successfully completing tasks. Businesses care about understanding their customers, managing risk, and accepting profitable transactions.

Identity is a means to those ends.

1. What Does “Identity” Actually Mean?

When most people hear “identity verification,” they think about proving who someone is in the real world: their name, address, date of birth, or government-issued identification.

But in technology, identity has a much broader meaning.

An identity can simply be an identifier associated with a cryptographic key, an account, or a software application.

There are five useful distinctions:

• Authentication: Can an entity prove it controls a particular credential?
• Pseudonymous identity: Can a business recognize the same user without knowing their real-world identity?
• Authorization: What has that entity been permitted to do?
• Identity verification: What real-world information about a person or business has been independently verified?
• Risk signals: What additional information helps a business determine whether an interaction is legitimate and safe?

These concepts overlap, but they’re not interchangeable.

An AI agent can be perfectly authenticated, authorized to access an account, and still be rightfully denied the ability to transact by a merchant.

2. The Technologies Behind Agentic Identity

Web Bot Auth: Which Agent Is Making This Request?

Web Bot Auth uses cryptographic HTTP signatures to authenticate automated web traffic.

Rather than relying on easily spoofed IP addresses or User-Agent strings, an agent signs its requests using a private key. A website can verify that signature against a corresponding public key.

This provides proof that the request was signed by the holder of that key. If the key is reliably associated with an AI platform, the website can authenticate the agent operator.

For example, a request might include:

Signature-Agent: "https://agent.example.com"

Signature: sig1=:BASE64_SIGNATURE:

Simplified illustration; not a valid signed request.

What Web Bot Auth doesn’t inherently establish is who the end user is or whether the agent has permission to act on their behalf.

PAP: Which Agent and Pseudonymous User Is This?

The recently announced Personal Agent Protocol (PAP) goes beyond authenticating requests.

PAP supports agent interactions with participating businesses, including sessions, pseudonymous user identification, and account authorization through mechanisms such as OAuth.

A particularly interesting feature is its use of stable, business-specific pseudonymous user identifiers.

For example, a PAP authentication assertion might contain:

{

}

Abbreviated example of PAP JWT claims.

The business can recognize the same pseudonymous user across sessions without knowing their real-world identity.

But this introduces an important trust consideration.

The agent platform is responsible for maintaining that identifier. If a user is banned for abusive behavior, the platform could technically issue a new pseudonymous identifier, making the same person appear to be a new customer.

Doing so would undermine PAP’s requirement for stable identifiers, but a business without another way to identify the customer may struggle to detect it.

Pseudonymous identity provides continuity only to the extent that the party assigning the identifier can be trusted to preserve it.

PAP and OAuth are particularly useful when a business already has an established customer relationship, authenticated accounts, and authorization infrastructure.

For example, a bank that has already verified its customers can extend existing account access to authorized AI agents without needing to independently establish the customer’s identity again.

That’s different from an agent arriving at an unfamiliar retailer where the customer has never created an account.

Identity Verification: Who Is the Actual Person?

Traditional identity verification providers such as Experian, Trulioo, Socure, and Persona establish real-world information about individuals and businesses.

Depending on the service, this can include legal names, addresses, dates of birth, government identification, business registration, and risk-related information.

This is what most people intuitively mean by identity verification.

These providers can establish who a person is, but don’t inherently establish which AI agent represents that person or what the agent is authorized to do.

KYA: Connecting Agents to Verified Real-World Identity

Know Your Agent (KYA) connects agent identity with information about the real-world people or businesses behind those agents.

At Skyfire, we’ve developed KYA to allow agents to present verifiable information about themselves, their operating platforms, and the users they represent.

For example, a simplified Skyfire KYA token payload might contain:

{

}

Abbreviated illustration using Skyfire KYA claim names. 

Rather than simply recognizing a pseudonymous identifier, the merchant can receive independently verified information about the real-world entity behind the agent, subject to the applicable verification and disclosure permissions.

This is especially valuable when the merchant has no existing relationship with the customer.

KYA answers a different question: Who does this agent actually represent, and what verified information can the business rely on?

3. Why These Differences Matter

For AI platforms, identity is ultimately about task completion.

An agent might authenticate successfully at a website’s edge, only to be blocked at login or rejected by a fraud provider during checkout.

Authentication success is not the same as task completion success.

For merchants, the priorities are different: understanding customers, preventing fraud, managing liability, and approving legitimate transactions.

According to Mastercard (a Skyfire KYA partner) and Datos Insights, the global financial impact of chargebacks is projected to reach $36.9 billion in 2026.

A merchant accepting a $15 purchase may need different assurances than one accepting a $5,000 transaction.

Knowing that an agent belongs to a recognized AI platform is useful. But it may not provide sufficient information for a merchant’s fraud systems to approve a high-value transaction.

This is also why existing customer relationships matter.

PAP and OAuth can be effective when a business already has verified customers, accounts, and authorization infrastructure.

Open-web commerce presents a different challenge. Many retailers support guest checkout, lack agent-accessible APIs, and rely on independent security providers across their websites, login systems, fraud screening, and checkout.

An agent may need to establish trust at multiple points before successfully completing a purchase.

The challenge isn’t simply enabling agent login. It’s enabling agent acceptance across the existing web.

4. Agentic Identity Isn’t a Winner-Take-All Competition

Web Bot Auth, PAP, OAuth, identity verification, and KYA don’t necessarily compete to solve the same problem.

They address different layers of identity and trust.

An agent could authenticate its requests through Web Bot Auth, establish a session through PAP, receive account permissions through OAuth, and present verified real-world identity information through KYA.

Different businesses will require different combinations.

A publisher serving public content may only need to authenticate an agent operator.

A bank serving an existing customer may prioritize delegated authorization.

A retailer accepting a high-value purchase from an unfamiliar customer may require additional verified identity and risk signals.

There is no reason every interaction should require the same level of identity assurance.

The goal should be to deliver the right information to the right business, at the right moment, with appropriate permissions and privacy protections.

5. The Future of Agentic Identity

As AI agents become more capable, the industry needs to move beyond simply asking whether a request came from an AI agent.

The more useful questions are:

Who operates this agent? Who does it represent? What is it authorized to do? What information has been verified? And does the business have enough confidence to accept the interaction?

Knowing which agent is making a request is not the same as knowing who that agent represents.

Knowing who it represents is not the same as knowing what it’s authorized to do.

And knowing all three doesn’t guarantee that a business will accept the transaction.

Ultimately, the goal isn’t simply to identify AI agents.

It’s to help legitimate agents successfully complete tasks while giving businesses the information, control, and confidence they need to welcome them.