Listen to today’s Open Channels News, your source for the latest news and developments in open source and artificial intelligence. Host Robert Jacobi gives us a report on today’s headlines explore the real meaning of openness in AI and the commercial power of open models.
Here’s what’s making news:
- OpenWALDO Launches to Redefine “Open” in AI: The founder of Rocky Linux, Gregory Kurtzer, has introduced OpenWALDO, a community-driven project aiming to bring true transparency to AI by ensuring the sources, licenses, and data origins of training materials are fully known and inspectable.
- AT&T Bets Big on Open Source AI: AT&T is rapidly shifting its AI infrastructure to open and open-weight models, already reaching 25% usage with expectations of climbing to 70–80%. The shift is driven by significant cost savings, AI sovereignty, and greater control over proprietary business data.
Stay tuned as Robert breaks down why the definition of “open” matters more than ever, and how these developments are shaping trust, control, and innovation in the AI industry.
Our sponsors keeps our network independent and actively publishing.
Since 2005, Automattic has built tools for the open web including WordPress.com, WooCommerce, and Jetpack that are used by millions of people to create, sell, and publish online. They believe in ownership, flexibility, and open source and we’re grateful for their support. Learn more at automattic.com.
In five days, Omnisend moves every flow, list, and template off your current platform, and you could be paying up to 35% less without lifting a finger. You just show up when it’s done. We’re glad to have them supporting the show and the community we’re building around it. Use the code OpenChannels and get 30% off your first 3 months of any paid plan.
Takeaways
OpenAI Models Aren’t Fully Open: Rob explains that many so-called “open” AI models only provide open weights, not full transparency into their data sources or the processes behind their creation, raising concerns about true openness and reproducibility.
The Open Waldo Initiative: Gregory Kurtzer, founder of Rocky Linux, is launching Open Waldo, aiming for true openness in AI by making training data, origins, licenses, and processes fully visible—an “AI bill of materials” to bring transparency and accountability to model foundations.
Importance of Data Provenance: Rob notes that organizations need to know exactly what’s in training data—its sources, licenses, and who contributed it—to mitigate legal and ethical risks, including copyright and copyleft license issues.
Model Collapse & Data Quality Risks: Rob highlights that training AI models on outputs from other AIs leads to “model collapse,” likened to making a photocopy of a photocopy, degrading data quality and transparency.
Community-Governed Open Data: Open Waldo proposes a community-managed shared dataset for AI, allowing the community to review, improve, or remove data once for all users, paralleling the way open source infrastructure works for software.
AT&T’s Shift to Open AI Models: AT&T is increasingly relying on open source AI, using open models for about 25% of its AI use (aiming for up to 80%) to control costs, enable customization, and avoid dependence on proprietary providers.
Cost and Control Drive Openness: For companies like AT&T, open models offer significant cost reductions—up to 80–90%—in large-scale AI deployment and the ability to run AIs internally for greater control over sensitive data and processes.
Emerging Concept of AI Sovereignty: Rob introduces “AI sovereignty,” the idea that companies want control over where their data goes and which models process it, resisting vendor lock-in and the risks of outsourcing critical infrastructure.
Open Source as Business Infrastructure: The podcast draws a parallel to previous open source successes like Linux and Apache, suggesting that a truly open foundation for AI could allow businesses to innovate on top of shared, transparent infrastructure rather than re-inventing basics.
The Underlying Commercial Demand: Rob emphasizes that enterprise interest in genuinely open AI models is driven by practical needs: cost savings, choice, trust, and the ability to switch providers without starting from scratch.
Mentioned Links and Resource
- Markets Insider (Source of the Open Waldo announcement) – The Open Waldo project’s official announcement was published via Markets Insider. 🔗 https://markets.businessinsider.com/
- The Wall Street Journal (Reporting on AT&T and open AI models) – AT&T’s shift towards open weight and open source AI was covered by The Wall Street Journal. 🔗 https://www.wsj.com/
Episode Transcript
Robert Jacobi:
Hello and welcome to OCN, the Open Channels News broadcast. I’m Robert Jacobi, your host. We lead today with open source and artificial intelligence, and a pretty basic question. When we call an AI model open, how open is it really?
Rocky Linux founder wants to make AI actually open. According to an announcement published by Markets Insider, the founder, Gregory Kurtzer, the founder behind Rocky Linux, CentOS, Warewulf, and Apptainer, is launching a new open source project called Open Waldo. And he’s taking aim at one of the stranger definitions we’ve developed during the AI boom. Open. Because a lot of what we call open AI today isn’t really open source. It’s open weight. And there is a difference, pretty big difference.
An open weight model lets you download the model weights, run them yourself, and potentially customize how the model operates. That’s useful. It’s certainly more open than a completely proprietary service. But you usually don’t get everything that created the model. You don’t necessarily know exactly what data trained it. You don’t necessarily know where that data came from, and certainly you don’t know whether that data was properly licensed. And you probably can’t reproduce the entire model from scratch.
OpenWaLDo wants to change that. The name stands for Open Weights, Artifacts, Licenses, Data, and Origins. Where’s Waldo? Which is admittedly a very open-source way to get to the acronym WaLDo. But the important part is the data. Kurtzer’s argument is that open source has already built most of the infrastructure — Linux, containers, Python, PyTorch, cloud infrastructure, supercomputing. All of that sits underneath the AI industry. But when we reach the training data itself, suddenly things get considerably less open.
Open Waldo is attempting to create a shared community-governed body of AI training data where the sources, licenses, and provenance of that data are actually known. Sort of think of, you know, that bill of materials that we’ve addressed in the last few episodes. If you’re running an application, increasingly you want to know what’s inside it, which libraries, which dependencies, which versions, where do they come from, any known vulnerabilities, all that fun stuff. Open Waldo wants to apply that same basic concept to artificial intelligence. Where did the training data come from? What license was attached to it? Who contributed it? What corpus was used for this particular training run? And ultimately, which data contributed to the model you’re actually using? That’s an AI bill of materials, ABOM.
And there is a practical reason this matters. Companies increasingly have to answer uncomfortable questions about the AI they’re deploying. Does the training data contain copyrighted material? Does it contain software under a copyleft license? Did somebody actually have permission to use the data. And here’s an increasingly weird problem: was the model trained on the output of another AI model? Because we’re beginning to create AI systems trained on data created by other AI systems that may themselves have been trained on data created by other AI systems. Fun recursion all around. It’s the digital version of making a photocopy of a photocopy of a photocopy. Eventually things get fuzzy, confusing and weird. Researchers call the AI version of that problem model collapse.
Open Waldo’s idea is to create a common foundation instead. Rather than every AI company independently collecting enormous amounts of foundational training data, the community builds a shared body of data once. Someone identifies a bad source, fix it once. Someone adds properly or improperly licensed data, everyone can benefit either from the fact that it’s good high-quality data or that it can be purged out of the system. Companies can then take that foundation, combine it with their own proprietary data, train their own models, and compete on what they add above it.
This should be disturbingly familiar. It’s basically how open-source infrastructure has worked for decades. Nobody needs to reinvent Linux before starting a software company. You start with Linux and build whatever makes your company different on top of it. Kurtzer wants AI training data to work the same way.
Open Waldo is being sponsored by CIQ, the company Kurtzer founded, but the project is intended to be community governed. Contributions will be public, attributable, and the data, tooling, and models produced by the project are intended to remain inspectable, for lack of a better term, and of course forkable.
Kurtzer summed up the argument pretty simply. Open source has won this argument before. The interesting question is whether it can win it again with AI. The story actually comes from an Open Waldo announcement published through Markets Insider.
For our next story, AT&T is betting big on OpenAI, and we don’t have to look very far to understand why this matters commercially. According to reporting from The Wall Street Journal, AT&T is making a major move toward open weight and open source artificial intelligence. Today, open models account for about 25% of AT&T’s overall AI usage. The company expects that number eventually to reach somewhere between 70% and 80%. It’s not an experiment anymore, it’s infrastructure.
And the reason is pretty simple: money. AT&T processes an average of about 45 million AI tokens every day. And at that scale, the difference between an expensive frontier model and a cheaper model that is good enough for the job becomes very, very real.
So AT&T built what it calls a smart router. A request comes in, the figure — system figures out what kind of AI capability it actually needs, then it routes that request to the most cost-effective model capable of doing the job. Not every question needs the smartest model on earth. If I need someone to drive me 3 blocks to the grocery store, I don’t need a Formula 1 car. I just need a car. AI is becoming the same thing. Use the expensive model when the problem actually requires it, use something smaller and cheaper when it doesn’t.
According to AT&T, switching some applications from proprietary models to open models has reduced costs by 80 to 90%. And when you’re processing 45 billion tokens a day, billion with a B, that’s not an optimization, that’s a business model.
AT&T now has more than 1,000 internal AI use cases, period. Open models are being used for things like legal and finance operations, customer service calls, helping field techs, and even managing parts of the actual AT&T network. The company has customized an open model using telecom-specific data to help agents identify the root cause of network problems. When I say agents, talking about AI, it’s always AI agents.
But cost isn’t the only reason AT&T likes open models. It’s the control. With an open model, AT&T can run the AI inside its own data centers. Corporate information doesn’t necessarily have to leave the company’s infrastructure and travel to an outside AI provider. That’s particularly important when the information involved is your confidential proprietary business data, customer information, network data, and of course intellectual property. And AT&T is calling this AI sovereignty, and I think that’s going to become a term that we start hearing at a minimum more and more.
Companies spent the last 20 years moving infrastructure into the cloud. AI may create an interesting counter-pressure, not because companies suddenly want to build everything themselves again, but because the model increasingly sits between the company and its most valuable information. And companies want control over that relationship.
There’s also the old-fashioned problem of vendor lock-in. AT&T, and honestly most folks, don’t want to become dependent on one AI provider. Open models give certainly AT&T, but all of us, many more options. If one model becomes too expensive, another model can potentially take its place. If a better model appears, the smart router can start sending appropriate workloads there. The app doesn’t necessarily have to care who’s underneath it.
And that’s where the 2 stories connect. AT&T is demonstrating the commercial demand. Open Waldo is trying to build the open source foundation underneath it. Enterprises want AI they can control. They want AI they can run themselves. They want to know what they’re paying. They want to know where their data goes. And increasingly, they want the ability to switch providers without rebuilding everything from scratch.
If we’re gonna call those models open, Kurtzer is asking the next logical question. Why stop at the weights? Why shouldn’t we know where the training data came from? Why shouldn’t licensing and provenance travel with the model? Why shouldn’t the entire foundation be inspectable, reviewable?
We’ve already seen this movie. Open source didn’t win because companies suddenly stopped wanting to make money. It won because sharing the commodity infrastructure underneath our products allowed everybody to spend more time competing on the things that actually mattered. Linux became infrastructure, Apache became infrastructure, Kubernetes became infrastructure, WordPress, maybe the foundational data behind AI becomes infrastructure too. And if that happens, the AI industry may discover something the software industry learned decades ago. You don’t have to own every layer of the stack to build something valuable on top of it.
That’s today’s edition of OCN, Open Channels News. Today we looked at 2 sides of the same OpenAI story. Open Waldo is asking whether OpenAI should mean more than downloadable model weights, and AT&T is showing why businesses increasingly care about openness in the first place: cost control, choice, and ultimately trust.
Thank you for joining us. For continuing coverage of open source hosting, cybersecurity, artificial intelligence, content management, and the domain name industry. Join us again tomorrow. I’m Robert Jacobi, your host. Be open and stay secure.
