Microsoft opens gated open-source AI models in Foundry as enterprise governance takes focus
Microsoft Foundry now supports gated models from Hugging Face, giving enterprises controlled access to open-source AI tools as demand grows for stronger governance, safety, and compliance.
Microsoft has expanded its AI model catalog in Microsoft Foundry to include gated open-source models from Hugging Face, allowing organizations to deploy advanced models within Azure while maintaining tighter controls over access and usage.
The update reflects a broader shift in enterprise AI adoption, as organizations look beyond model availability toward governance, accountability, and responsible deployment, particularly in regulated environments such as education, government, and child-focused digital platforms.
The change was highlighted in a LinkedIn post by Jessica Hawk, Corporate Vice President of Azure Product Marketing at Microsoft, who pointed to safety-driven use cases as a key driver.
From open access to governed deployment
Microsoft says gated models sit behind explicit access boundaries defined by model publishers, requiring users to request and receive approval before deployment. In Foundry, this process is integrated directly with Hugging Face user access tokens, allowing organizations to stay aligned with licensing and usage conditions while avoiding manual approval workflows.
According to Microsoft, this approach is intended to give enterprises access to a broader range of high-quality open-source models without sacrificing oversight. Gated models are presented alongside other models in the Foundry catalog, with indicators showing when additional permissions are required.
In her LinkedIn post, Hawk referenced Roblox as an example of why governed access matters at scale. She wrote that Roblox processes billions of chat messages and relies on a purpose-built PII Classifier model to flag personal information in real time, adding that the same model is now available through Microsoft Foundry.
Models rolling out first
Microsoft says gated models will be added to Foundry on a rolling basis, starting with a mix of computer vision, language, and safety-focused tools. Initial models include:
Segment Anything Model 3 (SAM 3) from Meta, which supports prompt-based object segmentation for use cases such as medical imaging, robotics, and content moderation.
Roblox PII Classifier, designed to detect personally identifiable information in chat messages using anonymized, multilingual training data.
FLUX.1 Schnell from Black Forest Labs, a text-to-image model optimized for fast, high-quality image generation.
EuroLLM-9B-Instruct, a multilingual language model supporting more than 30 European languages.
Bielik-11B-v3.0-Instruct, an instruction-tuned language model focused on multilingual reasoning, with particular strength in Polish and other European languages.
Microsoft notes that organizations can deploy these models directly into secure online endpoints within Foundry once access has been approved and verified.
For EdTech providers and education systems, the move highlights growing demand for AI tools that are both powerful and governable, especially when deployed in environments involving students, minors, or sensitive data. As AI skills development increasingly includes model selection, deployment, and oversight, platforms like Foundry are positioning governance as a core competency rather than an afterthought.
ETIH Innovation Awards 2026
The ETIH Innovation Awards 2026 are now open and recognize education technology organizations delivering measurable impact across K–12, higher education, and lifelong learning. The awards are open to entries from the UK, the Americas, and internationally, with submissions assessed on evidence of outcomes and real-world application.
