A hackathon that turned trustworthy AI ambition into working proof

I am still buzzing!

There are events where people talk about the future, and there are moments when an industry starts building it. At Innovate Americas, we created the second kind. For three intense days, 220 developers across 40 teams joined on-site in Dallas and virtually from around the world for TM Forum’s first Trustworthy AI & Data Hackathon.

The energy was extraordinary. We had to persuade developers to step away from their screens for the group photograph, and then they rushed straight back to coding. But that excitement matters because of what sat behind it: a real telecommunications challenge, a live autonomous enterprise sandbox and a shared determination to prove that AI can be governed without slowing innovation.

Not another proof of concept

The brief was deliberately practical. Teams ran TM Forum’s autonomous fault-resolution use cases on the AI-Native ODA Canvas. They configured Canvas and Kubernetes, connected components, ran agents, called models and engineered controls. This was not a theoretical discussion about responsible AI, but a hands-on validation by the people who must make autonomy work when they get back to the office.

By the finish, participants had created 196 additional agents and executed more than 35,000 governed model calls. Those agents monitored cost[AM1] , gathered evidence, controlled activity, connected tools and tested whether the environment could be trusted. In the process, the hackathon became one of the most substantial hands-on validations yet of the AI-Native ODA Canvas: 50 Canvas instances were running, and a new community of developers learned how to configure, integrate and build on ODA.

MODaaS moved from concept to operating control

Every model request passed through TM Forum’s Model Operations as a Service (MODaaS) implementation which was acting as the governing gateway for the hackathon. It managed policy, identity, authentication, encryption and guardrails, and dynamically routed model traffic inside the AWS environment. Across the challenge it handled the governed model traffic and over 350 million tokens with 10k to 11k tokens per call .

That is what I call adoption validation. MODaaS was not just sitting as a conceptual document but was being used as a real model, under sustained load, to govern real agent behavior and produce evidence of trustworthiness. What is more important, the environment also refused attempts to use undeclared models or bypass controls. Trust was not added after the event but was enforced at runtime.

The teams went beyond the challenge

The most exciting work came when participants started extending the environment. One team went underneath the infrastructure, extracted audit information and built an intrusion-detection capability for agents, connecting it to a compliance dashboard. The winning Infosys team, Infyvibes, demonstrated controls for spend, event recording and performance drift, then added real-time enforcement and a human-in-the-loop escalation when defined limits were breached.

This is why the hackathon felt like an industry changing event. It showed that governance can be coded, controls can be tested and evidence can be generated while autonomous systems are operating. It also showed what happens when TM Forum standards, architecture and industry expertise are placed directly in the hands of builders.

A new capability for TM Forum and the industry

The hackathon was hybrid by design, supported by Accenture, AWS, Dell Technologies, NVIDIA and ServiceNow, with AT&T hosting the first two days at its Dallas headquarters before the finale at Innovate Americas 2027 in Dallas. The operating model worked across locations, teams and time zones, supported through live collaboration, breakout sessions and expert guidance.

It also became the first deliberately orchestrated step in our ‘Trusted Decision and Action’ ambition work of the Governance Member Acceleration Program, which brings Accenture, AWS and ServiceNow together to focus on practical outcomes. Instead of beginning with another round of presentations, the program began with working code, evidence and lessons that can feed directly into industry delivery.

Now we scale this community and extend the trust

The teams have already told us what they want next: more network simulation and data, more model choice, and red-team challenges that actively test whether governance and evidence can survive attack. That is exactly the direction in which this should go.

So our job is to keep that community moving, bring the strongest contributions back into TM Forum collaboration, and turn what worked in Dallas into repeatable assets, methods and standards for the global industry.

So we are excited to announce we will be bringing an extended hackathon to TM Forum’s Accelerate event in Amsterdam in February 2027.