Nordic APIs Summit 2026 Recap: AI Agents, OAuth and API Security

October brought another edition of the Nordic APIs Summit. I always enjoy the event, not only for its merit, but also for the opportunity to visit Stockholm when autumn colors are at their most vibrant and the weather is still bearable. It’s a great time for sightseeing and Stockholm is, for sure, a beautiful city. Even better when you can combine that with an event like the NAPI Summit.

This year’s edition had a format similar to last year’s. We’ve kicked off with a half-day unconference, followed by two days of the main conference. The format offered plenty of  opportunities to learn from experts, practitioners and vendors about the current state of the API ecosystem.

Nordic APIs Unconference: From OAuth Adoption to Workload Identity

If you’re not familiar with the format, an unconference is a type of event where attendees gather to participate in discussions on topics they choose themselves. There is no predefined agenda. Participants propose topics and vote on those they find most interesting which then form the day’s agenda. Unlike regular conference talks, unconference sessions are discussions where everyone can share their insights, experiences, or questions.

This was the second time we’d held an unconference at the Nordic APIs Summit. The discussions showed that while AI agents and agentic access are definitely on everyone's minds, they’re far from the only topics the API community cares about. We also talked about scaling APIs, API quality, and standards.

As OAuth practitioners we tend to forget that there are still organizations that don't use it. We had a very interesting session about "how to switch to OAuth". It was a perfect example of what makes unconferences so valuable: the topic was proposed by someone who wants to collect knowledge from others, not share themselves. The session made me realise that knowing about a technology doesn't necessarily make adopting it straightforward. Taking the first step towards OAuth can be difficult and there are relatively few resources to help with that.

The unconference is always a good place to learn about new technologies and trends. Two particular interesting ones I learned about this year are:

• OpenID Workload Identity Security Events — a proposed specification which builds upon Shared Signals Framework and Security Event Tokens to standardize security events tied to state changes in workload identities.

• Dogwood, an extension to the Cedar policy language. It allows writing policies that take into consideration events that happened outside of the current request. For example, in the agentic world, it allows to deny a request to a “send email” tool if the agent previously read sensitive data (thus ensuring that the agent will not send that data to an unauthorized party). Such an approach to defining policies can potentially limit the amount of human approvals an agent needs.

Main Conference Highlights: AI Agents, API Security and Governance

Agentic was the theme of the year, for sure. I could even see some agentic topics that were present last year, but with a little twist. 

API design matters for AI agents

Both last year and this year, there’s been a lot of talk about improving API definitions to make APIs easier for agents to use. This year, however, there was also proof of the actual impact of API design. 

For example, Milton Carranza showed in his talk  - Your (Bad) API Design is Costing you Tokens - results of an experiment in which he measured the real impact of API design on coding agents. APIs with bad design required the agent to use more tokens and more iterations to successfully call an API. As Lorna Mitchell noted in her talk API Standards for AI Agents: “What's good for humans is good for agents.” If we focus on making our APIs easy to use for human integrators, then the agents will also consume them easily.

Securing AI agents through authorization and governance

Another recurring theme was the broader environment in which you run your AI agents. Natalia Venditto gave a perfect summary of this in The Agent Harness: Your New API Gateway: "An agent without control is just a demo". 

Quite a few speakers emphasized the  importance of ensuring the security and reliability of agents through all the elements around the agent itself, e.g.having a proper harness, a platform on which agents run, governance rules in place and clearly defined and enforced authorization policies.

There was considerable discussion about authorization and agents' privileges. It was encouraging to see that people care about this topic, as it is also something we care a lot about at Curity. And there is definitely a lot to discuss, as the community weighs different approaches to the matter, for example, whether we should treat agents as users or give them their own identities (different from users and applications). Or whether we should allow them to act in their own name or on behalf of users.

When talking about agents' privileges many people mentioned token exchange as a way of ensuring that the agent does exactly what it is supposed to do. I was glad to hear this as at Curity we also believe using token exchange is the correct way of solving the problem of authorization delegation in agents.

Another popular topic mentioned was the Shared Signals Framework; a specification from the OpenID Foundation that facilitates propagating security events in a system. It seems that the standard appears to be gaining importance as organizations look for ways to improve security for the agentic era

API standards and tools worth exploring

I always enjoy talks that leave me with a long list of things to check out or read up on. I wanted to highlight two sessions that I attended this year. 

The first was Andrzej Jarzyna’s Sovereign DevOps-Driven API Governance in Practice. Andrzej demonstrated a number of open source tools that, when used together, allow organizations build robust API governance systems, without having to rely on large and costly products. 

The second was from my Curity colleague, Judith Kahrer. In Cultivating the Right Standards for API Security as AI Agents Take Root she gave an excellent overview of the different standards already or soon-to-be used in the agentic environment.

Looking Ahead

Conferences are always a great place to spot industry’s current trends, and Nordic APIs Summit is no exception. This year, again, left me with things I need to read up on and ideas to try out. I can’t wait to see what the next year brings.

The conference recordings are now available on the Nordic APIs Youtube channel.