Licensing

Microsoft Product Terms Update: January 2026

print
share

Summary

Microsoft’s January 2026 Product Terms update formalises Security Copilot in M365 E5, rebrands Azure AI Foundry to Microsoft Foundry, adds new third-party AI model licensing terms, and clarifies Teams Phone carrier liability.

Microsoft’s January 2026 Product Terms update introduces four changes:

  • Azure AI Foundry rebranded to Microsoft Foundry, with third-party model licensing terms

  • Security Copilot inclusion in Microsoft 365 E5 formalised in Product Terms

  • Audio Services liability disclaimer for Direct Routing and Operator Connect

  • Dynamics 365 Business Central compliance certification for Spain

The changes are effective 1 January 2026.

Microsoft Foundry: Rebrand and Third-Party Model Terms

Microsoft has rebranded Azure AI Foundry to Microsoft Foundry. The name change appears throughout the Product Terms, Privacy and Security Terms, and Glossary. What was previously called “Azure AI Foundry Models” is now “Microsoft Foundry Models (includes Azure OpenAI)”.

Azure AI Foundry has been renamed to Microsoft Foundry in the January 2026 Product Terms.

This is the second rebrand in twelve months. Azure AI Studio became Azure AI Foundry at Ignite 2024, then Azure AI Foundry became Microsoft Foundry at Ignite 2025. Microsoft’s justification is that dropping “Azure” from the name signals agents are now “first-class citizens” in the Microsoft ecosystem rather than just another Azure service. Azure Active Directory became Microsoft Entra ID. Office became Microsoft 365. Microsoft 365 became Microsoft 365 Copilot. Each rebrand triggers documentation updates, training revisions, and support ticket confusion.

“Microsoft marketing is good at fails,” wrote one user on Microsoft’s own Q&A forum. "These awful naming decisions have caused confusion for our end users and support staff alike.” When Azure AD was renamed, The Register reported similar sentiment: “This change is both unnecessary and wasteful of customers’ time. MS should spend more time improving products and less time renaming them.”

If you’re writing procurement documents, updating internal wikis, or training staff on Microsoft’s AI platform, you’re now doing that work twice in twelve months. The platform underneath may be improved, but the name changes create busywork that benefits no one except, presumably, someone’s marketing dashboard.

So, the rebrand itself is cosmetic, but the January update includes something more substantive: licensing terms for Azure Direct Models. These are third-party AI models that Microsoft hosts and serves directly from Azure infrastructure, as opposed to models you deploy yourself or access through external APIs. The terms govern what you can and cannot do with models from Meta, xAI, DeepSeek, Black Forest Labs, and Mistral when accessed through Microsoft’s platform.

DeepSeek and MAI-DS-R1

DeepSeek-R1, DeepSeek-V3-0324, and MAI-DS-R1 are licensed under the MIT License. The MIT License is one of the most permissive open-source licences available, allowing commercial use, modification, and distribution with minimal restrictions. The main requirement is attribution: you must include the original copyright notice and licence text in any copies or substantial portions of the software.

For enterprises using these models through Azure, the MIT licensing means few commercial restrictions beyond what Microsoft’s own terms impose. You can use outputs commercially, modify how you interact with the models, and build products on top of them.

Grok

xAI’s Grok models are subject to xAI’s acceptable use policy in addition to Microsoft’s Product Terms. The acceptable use policy prohibits using Grok for illegal activities, generating harmful content, or attempting to circumvent safety measures. If you’re already compliant with Microsoft’s acceptable use policies, the xAI policy is unlikely to impose additional constraints for typical enterprise use cases.

Llama 3 and Llama 4

Meta’s Llama models come with more significant commercial restrictions that organisations need to understand before deploying them in production.

The first restriction is a monthly active user threshold. If your organisation has 700 million or more monthly active users across all products and services, you must request a separate licence directly from Meta before using Llama models. The threshold applies to your entire organisation’s user base, not just the application using Llama. For most enterprises the 700 million threshold is not a concern, but large consumer-facing platforms need to check their numbers.

The second restriction affects European deployments specifically. Multimodal versions of Llama, meaning versions that can process images, video, or audio alongside text, may not be used to provide services within the European Union. Text-only usage remains permitted. The EU restriction is a significant constraint for organisations building customer-facing AI applications in the EU that need image understanding or audio processing capabilities. If you’re deploying in Europe and need multimodal features, Llama is not currently an option for that functionality.

Multimodal Llama models may not be used to provide services within the European Union; text-only usage remains permitted.

The third restriction is an indemnification requirement. Customers must indemnify Meta against any third-party claims arising from use or distribution of Llama models. If someone sues Meta because of how your organisation used Llama, you’re on the hook for Meta’s legal costs and any damages. Indemnification clauses are common in software licensing, but they shift risk from the vendor to the customer, and your legal team should understand what they’re agreeing to.

The Product Terms also require attribution: if you use Llama models in a product or service, you must prominently display “Built with Llama” on your website, user interface, or documentation. You must also comply with Meta’s acceptable use policy, which prohibits using Llama for illegal activities, generating harmful content, or building competing AI models.

Black Forest Labs (FLUX)

Black Forest Labs provides image generation models through Azure Direct Models. The Product Terms include content filtering obligations: customers must either implement content filtering measures to prevent unlawful or infringing content, or ensure outputs undergo review before distribution. Customers must also implement “reasonable abuse monitoring measures” for end user activity.

The terms include broad indemnification requirements similar to Meta’s, plus a California Civil Code Section 1542 waiver. Section 1542 normally protects you from unknowingly releasing claims you didn’t know you had. Waiving it means you’re releasing Black Forest Labs from liability even for issues you aren’t currently aware of. California-specific legal provisions like this often extend to all users regardless of location when the vendor is California-based.

There’s also a competitive use restriction: outputs may not be used to create synthetic training data for AI models with “substantially similar functionality” to Black Forest Labs models.

Mistral

Mistral’s terms are more straightforward than the other third-party model providers. Mistral retains ownership of the models, but makes no claims to “Customer Developments” – customisations you make to the models or developments in furtherance of your use.

The indemnification terms are notably more favourable than Meta’s or Black Forest Labs’. Rather than requiring customers to indemnify the vendor, Mistral indemnifies customers against third-party claims alleging the Mistral models infringe patents, copyrights, trademarks, or trade secrets.

Security Copilot in Microsoft 365 E5

The January 2026 Product Terms formalise what Microsoft announced at Ignite 2025: Security Copilot is included with Microsoft 365 E5 subscriptions at no additional cost.

The new clause reads: “Each Microsoft 365 E5 subscription includes Security Copilot, with a monthly allocation of Security Compute Units (SCUs).”

SCUs are the billing unit for Security Copilot usage. Each query or action you perform in Security Copilot consumes a certain number of SCUs depending on the complexity of the request. Simple queries might use a fraction of an SCU; complex investigations that pull data from multiple security tools consume more.

The allocation formula provides 400 SCUs per month for every 1,000 licensed E5 users, up to a maximum of 10,000 SCUs per month. To reach the maximum allocation, you’d need 25,000 or more E5 users. If your usage exceeds the allocation, overage is billed at $6 per SCU through Azure on a pay-as-you-go basis.

For context, Security Copilot launched in April 2024 as a standalone Azure consumption service priced at $4 per SCU per hour for provisioned capacity. The E5 inclusion represents a meaningful addition for organisations already on E5, though the allocation may be modest for heavy security operations use. Organisations not on M365 E5 can still purchase Security Copilot standalone at the original consumption pricing.

The formalisation makes the Ignite commitment contractually binding.

Audio Services: Direct Routing and Operator Connect

Microsoft has added a liability disclaimer for Teams Phone System when used with third-party telecommunications carriers through Direct Routing or Operator Connect.

The new section confirms that Microsoft Phone System supports these integration methods, then establishes clear boundaries around responsibility. Authorised carriers control telephone number provisioning, not Microsoft.

Microsoft is not responsible for carrier billing. Payments for Direct Routing or Operator Connect services are managed directly between the customer and their chosen carrier.

The new section is a liability clarification rather than a licensing change. If something goes wrong with your phone numbers or carrier billing, Microsoft is making clear that you need to take it up with your carrier, not with Microsoft. The Phone System platform is Microsoft’s responsibility; everything that happens on the carrier side is not.

For organisations using Teams Phone with Direct Routing or Operator Connect, this doesn’t change how anything works.

Dynamics 365 Business Central: Spain Certification

Microsoft has added a compliance notice for Dynamics 365 Business Central in Spain. The notice states that Business Central complies with Spanish requirements for computer billing systems (sistemas informaticos de facturacion).

Spain has been implementing electronic invoicing requirements that mandate certified software for tax-relevant transactions. The Product Terms notice confirms Business Central meets these certification requirements, which matters for Spanish organisations evaluating ERP systems or existing Business Central customers who need to demonstrate compliance to auditors or tax authorities.

The Product Terms already contain similar regional compliance clauses, such as the Denmark bookkeeping certification.


🖐 Get clarity on Microsoft Product Terms and licensing changes. Learn more: Microsoft Licensing Services for Enterprises.


If you need help with Microsoft licensing, get in touch. We don’t sell Microsoft licences or cloud services, so our advice is independent.

Table of contents
print
share

Read next

More articles