The Short Answer
OpenWALDO, launched August 11, 2026 by Rocky Linux founder Gregory Kurtzer, makes AI training data auditable end to end — named sources, attached licenses, tracked provenance. ibl.ai is the agentic AI platform where you own all the code and the data, self-hosted inside your own perimeter, model-agnostic across any LLM, and usage-based with no per-seat pricing, so transparency reaches the whole stack.
An auditable corpus answers what went into the model. It does not answer where your data goes at inference time, or who controls the system asking the questions.
Those are separate layers, and a regulated buyer needs both to be inspectable.
What is OpenWALDO, and who is behind it?
CentOS and Rocky Linux creator Gregory Kurtzer just launched OpenWALDO — an open-source project for transparent AI training data you can audit end-to-end.
The pattern is familiar: proprietary lock-in creates opacity, and eventually an open alternative becomes the new standard.
The name is an acronym for Open Weights, Artifacts, Licenses, Data, Origins, and it launched on August 11, 2026, sponsored by Kurtzer's infrastructure company CIQ and structured as a founder-led, community-governed project.
Kurtzer's track record is the reason the launch registered beyond the open-source press. He founded Rocky Linux, co-founded CentOS, and created Warewulf and Apptainer — infrastructure that a large share of research and HPC computing runs on today.
The public corpus already indexes 124 billion reference tokens across 75.1 million documents, organized into 20 corpora and 1,051 shards, carrying 18 asserted license identifiers.
The design goal is that sources, licenses and provenance travel with the data from ingestion all the way through model release — so training data behaves like an open-source dependency: named, reviewable, versioned, attributable and verifiable.
Why does training-data provenance matter to an enterprise?
Because you cannot audit what you cannot see, and today most organizations depend on models whose inputs they have never inspected.
That opacity shows up in three concrete places, none of them philosophical.
Licensing exposure. If a model was trained on material whose license forbids commercial derivative use, the risk lands on the organization deploying it, not only on the lab that trained it.
Bias auditing. A fairness review that cannot reference the training distribution is describing symptoms. Provenance turns "the model behaves oddly for this population" into a question with a checkable answer.
Regulatory disclosure. The EU AI Act's general-purpose AI obligations require providers to publish a sufficiently detailed summary of the content used for training — a requirement that is difficult to satisfy retroactively when nothing recorded where the data came from.
Kurtzer's framing is that a licensed, provenance-tracked corpus is shared infrastructure — a common base layer that model builders extend with proprietary material, rather than each of them reassembling the basics in private.
What does the open-source precedent actually predict?
We've seen this movie before. Linux commoditized operating systems. Kubernetes commoditized container orchestration.
Each time, the open alternative won not just on ideology but on trust — organizations could inspect, audit, and modify what ran in their infrastructure. AI training data is following the same arc.
With OpenWALDO, the black box around training datasets starts to crack open. When you can see exactly what data shaped a model's behavior, compliance, bias auditing, and regulatory review all become tractable problems instead of hand-wavy assurances.
The question was never whether open alternatives would emerge — it was when.
It is worth being precise about what the precedent predicts, though. Linux did not win because it was free; it won because organizations could run it themselves, fix it themselves, and never be told what version they were allowed to have.
Transparency was necessary but not sufficient. Control was the payoff.
Where does auditability stop if you rent the platform?
At the model boundary — which is the part of this story most coverage misses.
An auditable corpus tells you what shaped the model's behavior. It says nothing about what happens to the records your organization sends through that model in production.
| Layer | Question it answers | Made auditable by |
|---|---|---|
| Training data | What went into the model? | Open corpora like OpenWALDO |
| Model weights | Can you run and pin it yourself? | Open-weight releases |
| Serving platform | Where does your data go, and who sees it? | Owning the source and hosting it |
A fully transparent model served through an opaque platform still leaves the most operationally sensitive question unanswered: what your organization sent, and who could read it.
How do you get transparency across the whole stack?
By owning the layer that the model runs inside. With ibl.ai you own all the code and the data — the full source under a perpetual license, on infrastructure you control.
The platform is model-agnostic, so a model trained on an openly-licensed corpus and a hosted frontier model are both routing choices, and provenance-conscious procurement stays possible as the ecosystem matures.
Because inference happens inside your perimeter, the audit trail for what your organization asked and what came back is yours by default — not a report you request from a vendor.
Pricing is usage-based with no per-seat pricing, and deployment runs anywhere: your cloud, your VPC, on-premise, GovCloud, or fully air-gapped.
More than 1.6M users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.
ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.
OpenWALDO is opening the bottom of the stack. Owning the top of it is what turns that transparency into something your auditors can use.