The practical challenge is that sovereignty spans a layered supply chain. Even when an internal team uses an approved tool, the organization may still inherit fourth-party and fifth-party risk through hosting, model providers, data brokers, and embedded services. That makes vendor management central. Senior leaders should push for contractual clarity on data retention, cross-border processing, audit access, model change notification, and exit options, rather than treating AI procurement as an extension of standard SaaS buying.
There is also an operating-model implication: centralized policy with distributed execution usually fails unless usage is visible. Inventory, logging, and monitoring are not administrative overhead; they are the minimum controls that make sovereignty enforceable. The same applies to shadow AI, where the biggest issue is often not malicious intent but unmanaged experimentation beyond approved boundaries.
Useful questions for leadership teams:
- Which AI use cases require sovereign-by-design controls, and which can tolerate shared infrastructure?
- Do we know where prompts, model outputs, telemetry, and training data actually travel?
- Can we produce auditable evidence of policy enforcement across internal teams and external providers?
- Is our incident response plan explicit about AI-related data leakage, model compromise, and regulator notification?
The trade-off is clear: tighter control can slow deployment and narrow vendor choice, but weak control can undermine trust, compliance, and negotiating power. The right target is not maximum isolation; it is deliberate, provable control over the AI stack elements that matter most to the business.
Nation states crave domestic dominance in the AI race, and while some vendors rush to offer sovereign AI infrastructure to their government clients, other enterprises grapple with prioritizing their organizations’ ambitions.
“It’s about maintaining control of your AI environment,” said Veena Dandapani, COO at identity authentication company SLC Digital. “It’s not just about the data that you’re feeding into the AI.” That includes thinking about the models used and where the model infrastructure is hosted, which makes jurisdiction a factor, she said.
What is at stake for C-suite tech leaders is control over AI’s development, infrastructure and supply chain — does it reside with a country’s policies or the enterprise?
The definition of sovereign AI is muddled because of differing perspectives and market objectives. IBM defined AI sovereignty as an organization’s or nation’s ability to control its AI tech stack, including the IT infrastructure, data, and models.
A report released in August found that 52% of the 508 IT and business decision-makers surveyed described sovereign AI “in terms of local/national control.” Market intelligence firm IDC conducted the survey, which enterprise AI company Cohere commissioned.
Another 35% of respondents referred to sovereign AI in relation to digital sovereignty and independence. Just 13% of respondents said sovereign AI is “widely or very widely” understood in their organizations, and one-third of respondents said they had difficulty describing the concept in their own words.
While organizations may not be entirely clear on AI sovereignty as a concept, they are aware of the risks of ignoring it. Data leakage, privacy breaches and compliance issues top the list of concerns in the Cohere and IDC report.
As CIOs lead AI strategies at their organizations, how do they approach sovereignty? InformationWeek spoke to three enterprise leaders about what they are doing about the AI sovereignty conundrum.
AI sovereignty: Data, infrastructure, models, operations
For Dandapani, AI sovereignty boils down to control. Achieving AI sovereignty requires an organization to make choices on its level of interdependence, according to a brief from the Stanford Institute for Human-Centered AI (HAI). Developing and operating an AI tech stack must be done with partners rather than in isolation. As CIOs choose and work with those partners — frontier model providers or otherwise — sovereignty is about strategic control and the flexibility to make changes.
On a global scale, government leaders are concerned about their countries’ overreliance on a small number of foreign vendors. “The dominance of a few large cloud providers — especially in the United States and China — has created fears abroad over vendor lock-in and exposure to foreign surveillance or political leverage,” according to the Stanford HAI brief.
Vendor lock-in also raises concern over autonomy, the heart of AI sovereignty, at the enterprise level. But the risks don’t end there.
More than two-thirds of leaders surveyed by Cohere and IDC see data leakage and privacy breaches as the biggest drivers of sovereign AI; more than half pointed to compliance, regulatory or legal risks.
The costly consequences of data leaks and compliance failures are well known. Enterprises can face direct financial damage and brand damage.
“You can really create a lot of problems, not only just from a regulatory perspective and cash perspective but also from a trust perspective,” Randy Dougherty, CIO at cybersecurity company Trellix, said.
How CIOs are approaching AI sovereignty
While the risks are coming into focus, CIOs are playing catch up on AI sovereignty. The initial drive, as is often the case with new technology, is on time to market, according to Dougherty.
“No one wants to be the uncool kid not using the technology. We want to take advantage of it, but often we don’t start with a secure foundation,” he said.
But that may change. Last year, IDC predicted sovereignty demands will drive CIOs of multinational organizations to up investments in “modular, sovereign-ready cloud and data localization environments” by 65% by 2028.
Christopher Morton, CIO at IT and managed service provider Logically, said he grapples with sovereignty challenges as his company becomes more AI-forward. “Trying to really manage the sovereignty of where is our data, who has access and how are we managing it … has become a much greater priority for us in the last quarter or so,” he shared.
Data can be messy, especially when enterprises work with a complex network of vendors. Many companies operate in multiple jurisdictions, which means different sets of regulations. Having one set of rules to adhere to across all borders would be easier naturally. “The truth is that’s not the world we live in. It never has been,” Dougherty said.
CIOs must be cognizant of where each piece of the AI tech stack is and how data moves through it. Is the enterprise subject to the EU AI Act? GDPR? Data residency regulations in the Middle East? Each jurisdiction has its own rules and regulations.
At Logically, the approach is to comply with “the most restrictive standards,” according to Morton. “We’re trying to align ourselves with NIST and really use that as the guiding principles,” he said. “We know … that in some areas we’ll have to bend and make accommodations and document appropriately.”
AI sovereignty becomes more complicated when CIOs examine vendor relationships. Are they buying third-party data and AI models? Where does that data come from? Where do those models and their underlying infrastructure sit?
CIOs and other technology leaders must think about the entire chain of custody for data, according to Dandapani. “Not only the source, but also usage all the way down to the fourth, fifth party,” she said.
In addition to these issues, the threat of shadow AI also looms large over sovereignty. CIOs cannot control AI they don’t know is in use within their enterprises. Morton believes there is “a miscalculation of just how prevalent it is” in many organizations.
So, how are CIOs and other enterprise leaders tackling the challenges surrounding AI sovereignty?
-
Creating an inventory. To maintain the control necessary for sovereign AI, CIOs need to know what AI resources are in use and how they are using enterprise data. Easier said than done. “It’s one of those things that’s very simple in policy to write down,” Morton said. “When the rubber hits the road, it does become quite difficult in an AI-enabled organization to thoroughly catalog all those various connections and automations.”
Initially, Morton and his team used a manual process to catalog automations, but it quickly became clear that wasn’t enough to keep up with the pace of change. Instead, they are developing an agent to catalog existing automations and who has access to them. -
Implementing data and usage controls. Sovereign AI requires enterprise-wide data and usage controls. Morton, for one, is focusing on wrapping a control layer around AI initiatives at Logically.
“We’re able to basically send all the AI traffic through that control layer and then we can wrap our governance around it,” he said. “We can do things like log certain prompts. We can flag things if there has been sensitive client information uploaded.” With all the proliferation of AI and the data flowing through that tech stack, CIOs need to be alerted to any data issues and improper usage. Dandapani described controls that continuously monitor for any anomalies. “Anytime we have any events that break policy, [our] systems catch it,” she said. These controls need to be extended to third-party vendors. Dougherty ensures supply chain compliance frameworks are enforced. “We have contractual and technical vetting for any third-party AI, ensuring things like zero data retention, and more importantly, that there’s no cross-border telemetry and exfiltration that can happen that would allow the data to leak anywhere,” he said. -
Updating incident response. As enterprises continue to push forward with AI, it is possible that their approach to AI sovereignty will fail. Data could be leaked. Compliance issues in different jurisdictions could arise. Enterprises need incident response planning for these scenarios. At SLC Digital, the incident response team is prepared to act if an AI model or the data feeding it is compromised. “We will immediately stop infusing any further data. Corrective teams come into place and start doing corrective action,” Dandapani said. “We will inform customers. We will also inform appropriate authorities that there is an issue with our models.”
-
Prioritizing auditability. One of the biggest AI sovereignty challenges comes down to access, according to Dandapani. Who accessed the data and the AI tool? How was a decision involving AI made? “Auditability is the most important. When an organization can produce evidence and be able to be transparent, they will be ahead of the game,” she said.
AI sovereignty requires ongoing conversations among enterprise leaders and continuous monitoring of the entire value chain, from data to models and infrastructure. Dougherty cautioned against regarding it as an afterthought. “The biggest mistake organizations make is that they treat AI sovereignty like a compliance checkbox rather than what I would consider a core survival strategy,” he said.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.