Feature
Research Report
Mind The Agentic Action Gap: Stop Losing Money With AI Agents
Open-source AI promises something many enterprises want from artificial intelligence: more flexibility, less vendor lock-in and more room to customize models for specific business needs.
In theory, it also creates a more dynamic ecosystem. Instead of relying on a single company to maintain, improve and govern a model or related software, open-source AI allows a broader community of developers, researchers and organizations to test, refine and improve the systems over time.
At its simplest, open-source AI refers to AI systems that developers and users can use, study, modify and share without needing permission from rights holders. That openness can accelerate innovation by allowing more contributors to participate across the training, fine-tuning and inference phases.
“This transparency provides the opportunity to continually advance the trustworthiness of a model and the software components that create and serve the model,” said Stephen Watt, distinguished engineer and vice president, office of the CTO, for Red Hat. A broad community, he added, can collectively identify and address vulnerabilities more effectively.
Table of Contents
- What Open-Source AI Actually Requires
- Why Transparency Matters for Trust
- Open Source as an Innovation Engine
- The Adoption Barriers: Clarity, Legal Risk and Infrastructure
- Why Provenance Is Becoming Critical
- Open Models Are Gaining Ground
What Open-Source AI Actually Requires
The definition of open-source AI has become increasingly important as more companies label models “open” without necessarily making the full system transparent.
The Open Source Initiative ran a global consultation in 2024 to clarify what open source means in the context of AI. The result, published as the Open Source AI Definition v1.0, states that machine learning systems should make several components available: trained weights, complete training code, code used to create training datasets and the training data itself unless legal restrictions prevent distribution.
From the perspective of Stefano Maffulli, executive director of the Open Source Initiative, open-source AI is about transparency across the whole stack: model architecture, training data and weights, all distributed under licenses that allow innovation without additional permission from rights holders.
“The idea is that anyone with the right skills should be able to understand, replicate and build on the system,” he said. “The benefits are huge.”
Those benefits mirror the traditional promise of open-source software: more transparency, greater trust, better science and faster technological progress, because developers do not have to constantly reinvent the wheel.
Related Article: Open-Source vs Closed-Source AI: Which Model Should Your Enterprise Trust?
Why Transparency Matters for Trust
Trust is one of the biggest arguments in favor of open-source AI. When organizations can inspect how a model was built, what tools were used and what data shaped it, they have a better chance of understanding its limitations and risks.
“Building trust in open source AI, especially with rapidly evolving open source models, requires a strong foundation of security, safety and transparency,” Watt said.
Organizations can start by adopting models from verifiable sources with documented information about intended use, evaluation data and governance. Software supply chain development best practices can also help organizations understand risks introduced during model development. Clear access control and enforcement, such as mutual authentication and zero trust practices, can further reduce the probability of model misuse.
“Understanding what tools and frameworks were used in the development and training of the model can increase transparency in what security and safety practices were applied and provide organizations insights into the model’s origin and potential risks,” explained Watt.
For Maffulli, transparency also helps level the playing field. When practitioners can look under the hood, it becomes easier to identify bias, errors or misuse. It also allows a wider range of people and organizations to do meaningful work with AI. "It speeds up innovation,” he said. “Open ecosystems tend to move faster because more people are experimenting, sharing, and improving things together.”
Open Source as an Innovation Engine
Open source has long been a powerful engine for software innovation. Open-source AI carries the same potential.
When models, data and related tools are open, researchers, startups and developers can build on each other’s work instead of starting from scratch. That can help the field move faster and address a wider range of problems.
“That’s how we move faster and solve more diverse problems,” Maffulli said. Of course, he added, there are tradeoffs.
One tradeoff is control. Centralized systems can make safety and alignment efforts easier to coordinate. But, Maffulli argued, over the long term, openness can create greater resilience if the ecosystem has clear licensing, transparent data practices and community accountability.
“It's not about no control; it's about shared responsibility,” he said.
Watt also pointed to the role open source plays in accelerating AI development. Open-source artifacts are easy to publish, discover and refine, which allows the broader ecosystem to build on them quickly. “The access, agency, diversity and trustworthiness of the open source way all contribute to its ability to accelerate innovation in AI."
The Adoption Barriers: Clarity, Legal Risk and Infrastructure
Despite the promise, open-source AI adoption still faces major hurdles.
The first is clarity. Many projects claim to be open, but Maffulli said that label can be misleading. “A lot of projects claim to be ‘open’, but if they don’t share the training data or weights, or if they use restrictive licenses, it’s not really open source in practice,” he said.
The second hurdle is legal risk. Without clear data provenance and licensing, organizations may worry they could face lawsuits if they use or distribute a model.
That creates a paradox, noted Maffulli. The more transparent models are often the ones sued more quickly, like the models from Eleuther AI. On the other side, models trained with careful analysis of copyright status may have less data available, which can lower performance.
Webinar
May
12
How Marketing Teams Can Take Control of the Web Experience
See how Veo and Kvalifik put marketing in the driver's seat, shipping faster and growing qualified bookings by 150%.
Register
Webinar
May
14
Two Audiences, One WordPress. Is Your Site Ready?
The practical steps to making your WordPress portfolio ready for both human visitors and AI agents.
Register
Webinar
May
19
From Content Sprawl to Competitive Advantage: A Knowledge Operations Roadmap
How automation, AI and a centralized content platform turn knowledge fragmentation into a scalable competitive advantage.
Register
Webinar
May
20
From AI Investment to CX Results: What Enterprise Leaders Need to Know
Move beyond experiments. See how top enterprises scale AI for CX results.
Register
Webinar
May
20
AI With Intent: How CX Leaders Decide Where Automation Belongs (And Where It Doesn't)
Senior CX leaders from REI & Hyatt reveal their framework for using AI to boost retention.
Register
Webinar
Jun
16
From Legacy to Launch-Ready: How Gainbridge Made Its Website a Marketing-Led Growth Engine
Join in to learn how a D2C annuity brand gave marketing full website ownership — without slowing down or risking compliance.
Register
“If you look at the problem with the other side, the models trained with careful analysis of the copyright status of content used in the training dataset are the ones with less data and therefore lower performance,” he said.
The third barrier is infrastructure. Open-source AI can require significant compute power, storage and technical expertise.
“If we want to make adoption easier, we’ll need better tools, shared resources and clearer governance around how open models are built, shared and trusted,” Maffulli said.
Related Article: Healthcare's AI Crossroads: Open Source or Commercial Foundation Models?
Why Provenance Is Becoming Critical
As open models become more powerful and more widely used, provenance is becoming a central issue for enterprise adoption.
Organizations need to understand who built a model, how it was developed and whether using it could create business risk or legal exposure.
“Provenance is important because you have to understand the trustworthiness of the person who built the model and assess whether using the model could create any business risk or legal exposure for you,” Watt said.
The open-source ecosystem has been evolving to provide better capabilities around attestation and measuring trust. That matters because enterprises do not just need access to models. They need confidence that those models can be used safely, legally and reliably.
Open-source projects also rely on a variety of governance structures that benefit both contributors and users. That governance layer can help communities maintain trust while continuing to innovate.
Open Models Are Gaining Ground
The open-source AI debate is also changing because open models are becoming more capable.
Watt said the arrival of models such as DeepSeek and the Llama family has contributed to a shift in which open-source generative models are now outperforming proprietary models in some areas. “We haven’t seen any clear functional or non-functional inhibitors in the open models or the model architectures behind them."
That shift strengthens the case for open-source AI. If open models can deliver competitive performance while giving organizations more transparency, flexibility and control, they become harder for enterprises to ignore.
The case for open-source AI is not that openness solves every problem. It does not eliminate legal risk, infrastructure demands or governance challenges. But it does offer a different path for AI development — one built on transparency, shared responsibility and broader participation.
For enterprises, that may be the real value: a clearer view into how those models work, where their risks come from and how the ecosystem around them can improve.