Microsoft adds European access oversight and offline AI to its sovereign cloud

A sovereignty push launched in Paris

Microsoft announced new sovereign cloud features on 20 October at the Paris stop of its Frontier Tour, a series of touring events on digital transformation. Satya Nadella spoke from the stage at the Cnit centre in La Défense and pitched the company's AI strategy to customers, among them Stellantis, Club Med and Saint-Gobain, which were cited during the talks. The stated goal is to help organisations gauge their level of digital sovereignty and speed up AI adoption while keeping data protected and access tightly governed. The offer is aimed mainly at public bodies, hospitals and large companies that want to reduce their dependence on American and Chinese technology. Microsoft first launched its Sovereign Cloud in June 2025, at a time when Europeans increasingly worry about relying on American cloud providers.

Oversight of engineers and customer-held keys

Two new tools cover customers in the EU and in the European Free Trade Association, which includes Switzerland, Norway, Iceland and Liechtenstein. Under Data Guardian, a trusted European partner oversees the remote logins that Microsoft engineers make into customers' production systems. Production Touch Ledger keeps a record of that access that cannot be altered without detection. Customers can also keep the master key that protects their data on hardware they own, instead of entrusting it to Microsoft. Douglas Phillips, who leads Microsoft's specialised clouds unit, set out the logic behind the plan.

“Our approach to digital sovereignty starts with the workload.”

A new Sovereign Control Panel in the Azure portal will display the status of each system and flag gaps. It enters public preview at Ignite, which runs from 17 to 19 November in San Francisco.

Offline AI on customer hardware

Most of the new tools run on Azure Local, a Microsoft cloud build that sits inside a customer's own data centre. Foundry Local, which runs AI models on site, and GitHub Enterprise Local, which lets developers pair GitHub Copilot with locally hosted models, become generally available at Ignite. Both can work fully offline. SQL Server on Azure Local is already generally available. Microsoft 365 Local, which connects to a customer's own SharePoint and Exchange servers, is in preview for approved customers, and systems will be able to switch between connected and disconnected modes. Microsoft named 29 partners for the effort, among them Bleu in France, Delos Cloud in Germany, Mistral, Nvidia and Dell.

Agents treated as insider threats

In a parallel message, Microsoft argues that autonomous AI agents should be handled like insiders with access to company systems. Nadim Abdo, corporate vice president for identity and network access, wrote in a company blog that agents are not inherently malicious but can make mistakes or be manipulated. Abdo added:

“We should not ask a model to be its own security authority.”

Hayete Gallot, who leads Microsoft's security business, wrote that the most advanced models cannot be trusted whatever their origin, and that many agents receive more privileges than they need. Reports say the caution extends to models from partners OpenAI and Anthropic. Nadella had written on X on 10 October that systems should be designed on the assumption that a model may be compromised, with an authorised person able to stop it at any moment. Microsoft presents Guardian Agents as an extension of Project Perception, a system in which red agents probe for weaknesses, blue agents investigate threats and green agents implement fixes. Gallot warned that agents may act unpredictably, adapt as context shifts and coordinate with other agents toward their goals, even without any attacker involved.