At this year's FIDO Alliance Authenticate conference, one theme kept coming up across discussions about digital identity, payments, and AI agents: proving who someone is is only part of the challenge. We also need to prove what they're authorized to do.
Our CEO, Nick Lambert, attended the three-day conference, and we've put together six of the most interesting developments and insights he took away.
1. Verifiable credentials could learn a lot from passkeys
One of the lessons from passkey adoption is that people don't need to understand the technology to benefit from it.
What matters is delivering a better user experience, using familiar terminology, and giving people a compelling reason to switch.
It's a valuable lesson for the verifiable credentials industry, where adoption depends on more than the underlying technology.
2. Mobile credentials are moving beyond pilots
There were some interesting examples of digital credentials being used in the real world:
- Arizona is rolling out digital birth credentials statewide.
- France is testing digital travel credentials at a live border.
- Cross-border interoperability between mobile driver's licenses (mDLs) was demonstrated.
Another interesting development concerns ISO 18013-5, the standard behind mDLs. Server retrieval is expected to be removed from the core standard in its second edition and addressed in a separate technical specification.
Telecom operators also featured in the identity discussions. CAMARA, GSMA, and Google highlighted how mobile network capabilities can support authentication and phone number verification.
Given the trusted relationships operators already have with billions of customers, there's an opportunity to make that trust useful beyond their own networks.
3. Digital credentials are becoming part of payment infrastructure
Mastercard and PayPal both presented approaches to using digital credentials in payments.
Mastercard demonstrated a device-bound digital payment credential that could represent a payment card without exposing the card number or other sensitive information.
PayPal discussed how EUDI wallets could support Strong Customer Authentication (SCA) under PSD2.
An important distinction is that a Personal Identification Data (PID) credential alone isn't sufficient for SCA. Banks and payment providers need their own authentication credentials, bound to the user's device.
The WE BUILD pilots are exploring how EUDI wallets could hold these credentials, with work covering both card-level and account-level payment authentication.
However, there are still challenges.
OneSpan and Venable examined EUDI wallets against Strong Customer Authentication requirements. While authentication and transaction-signing capabilities look promising, the architecture they assessed doesn't provide banks with the device and behavioural signals they normally use for transaction monitoring.
This highlights an important tension between privacy-preserving identity and fraud detection.
EUDI wallets are designed to protect privacy and minimize unnecessary data sharing. But banks also need enough information to identify suspicious transactions, even when authentication itself is successful.
4. Agentic commerce is moving toward standardized protocols
Many AI shopping agents still operate through browser automation, logging into websites as users and navigating checkout processes.
That's changing as new protocols emerge:
- ACP: Standardized agentic checkout.
- UCP: Commerce workflows covering discovery through checkout.
- AP2: Cryptographic proof of what a customer authorized an agent to do.
An important detail is that Google's AP2 uses verifiable credentials (SD-JWT VCs) for its mandates.
Google has contributed AP2 to the FIDO Alliance, while Mastercard has contributed its Verifiable Intent framework. The two approaches are being harmonized through FIDO's Payments Technical Working Group.
It's encouraging to see work toward common standards for agentic payments, with verifiable credentials playing a role.
5. A valid agent identity doesn't mean an authorized action
This was one of the most interesting discussions throughout the conference.
Capital One gave an example of an AI agent with a valid authentication token, the correct permissions, and a recently authenticated user.
Despite all those security checks passing, the agent could still initiate an unauthorized transaction because of malicious instructions hidden in a document.
The takeaway was that authorization needs to be tied to the specific intent, with clear limits and verifiable evidence of what was approved.
A related question came up in research presented by Víctor Vargas (Microsoft).
Imagine an AI assistant has permission to answer emails and make payments.
The user asks it to answer emails, and one of those emails requests a payment.
Should the agent be allowed to make that payment simply because it has payment permissions?
This raises an important distinction between having permission to perform an action and being authorized to perform it in a particular context.
6. Verifiable delegation is becoming central to agent trust
Mastercard presented a layered approach to Verifiable Intent that connects three pieces of signed evidence:
- Identity: Proof of the verified cardholder, signed by the credential issuer.
- Delegation: The user's limits and authorized agent, signed using the user's bound key.
- Purchase: Details of the cart and merchant, signed by the agent.
Together, these form a verifiable chain connecting the human to the purchase.
PayPal highlighted another important issue: existing agentic payment protocols don't fully establish whose authority an agent is acting under.
An email address, wallet address, or payment card doesn't necessarily establish the identity of the person who granted permission.
PayPal proposed using verified identity credentials, such as mDLs or KYC credentials, to anchor delegation to a verified person.
Outside payments, Twilio and Anthropic discussed work on MCP authorization, including centrally managed permissions, sender-constrained tokens, workload identity, and identifying individual agents. Cedar also came up repeatedly as a language for expressing agent permissions.
An open question is how agents can prove their authority when acting across organizations that don't already share an identity or authorization system. This is where we believe verifiable credentials have an important role to play.
The bigger picture
If we had to summarize the conference in one observation, it would be this:
Digital identity, payments, and agent authorization are starting to converge.
We're making progress on proving who people and agents are.
But the next challenge is establishing who authorized an action, what limits apply, and whether the action respects those limits.
And as AI agents increasingly act across different systems and organizations, being able to verify that authority independently becomes particularly interesting.
That's one of the developments we'll be following most closely.
