Two reports released within weeks of each other in 2026 tell a story enterprises can no longer treat as a future risk. It’s a present one, and it’s getting worse quickly.
IBM’s 2026 Cost of a Data Breach report, built with the Ponemon Institute, found that one in five organizations reported an AI-related breach — and 92% of those breaches happened at organizations without proper AI access controls in place (Cybersecurity Insiders). The global average cost of a data breach is now $4.99 million, up 12% year over year, and breaches involving AI carry a meaningful surcharge: roughly $1 million in additional cost per incident. The most expensive categories are specific and structural — model inversion incidents average $6.07 million per breach, and prompt injection incidents average $5.89 million.
Separately, data from the Identity Theft Resource Center shows breach volume in the first half of 2026 already running ahead of last year’s record pace: 1,803 reported data compromises through H1, compared with 1,732 in the same period of 2025, associated with more than 471 million victim notices (CNBC). Between March 2025 and February 2026, one in four breaches was AI-enabled — a 56% increase from the year before. ITRC president James Lee summed it up bluntly: “We continue to see this ever-increasing number of data breaches. That does not appear to be slowing down.”
The uncomfortable detail buried in the IBM data
The most important number in the IBM report isn’t the breach cost — it’s the access control gap underneath it. Fewer than half of organizations are actively securing non-human identities, and only 40% control access to their AI models and data in any structured way. Put plainly: most enterprises have deployed AI systems with broad, poorly governed access to their data, and the breach data shows exactly what happens next.
This isn’t a failure of AI models. It’s a failure of the data layer those models sit on top of. Model inversion and prompt injection — the two most expensive breach categories in the report — are both fundamentally about an AI system’s relationship to the data it can reach. If an AI agent or model has more access than it needs, for longer than it needs it, without governance that follows the data rather than just the perimeter, the breach math in this report is the predictable result.
Encouragingly, the report also shows enterprises are starting to respond: more than half of breached organizations plan to purchase AI security and governance tools, an 88% increase in planned purchases year over year, and 85% intend to increase overall security spending. The intent is there. The question is whether that spending goes toward bolting more tools onto an ungoverned data foundation, or toward fixing the foundation itself.
Data residency adds a second layer of urgency
At the same time enterprises are confronting AI-driven breach risk, they’re facing a second, related pressure: where data is allowed to live at all. Gartner forecasts worldwide sovereign cloud infrastructure spending will reach $80 billion in 2026, a 35.6% increase from 2025, driven by regulatory frameworks like the EU Data Act — whose Chapter VII protections for non-personal data have applied since September 2025 and reach any provider serving customers in the Union, regardless of where that provider is based.
Security and sovereignty are converging on the same requirement: enterprises need to know precisely where their data lives, exactly who and what can access it, and be able to prove both, at any moment, across every environment they operate in. A cloud-only architecture makes that harder than it needs to be — every cross-border query, every AI agent reaching into a dataset, every access grant becomes a compliance and security question simultaneously.
Security-by-architecture, not security-by-add-on
This is the case for building AI and analytics on a platform where governance, access control, and data residency are architectural properties, not aftermarket additions. A hybrid, high-performance SQL data platform that can run securely on-premises, in the cloud, or across both gives enterprises a way to keep sensitive data exactly where regulation or risk tolerance requires it, while still giving AI systems and agents governed, auditable access to the data they need to do their jobs.
The breach numbers are a warning about what happens when that governance is missing. The sovereignty numbers are a warning about what happens when the infrastructure choice itself becomes a compliance liability. Enterprises serious about scaling AI in 2026 need a data foundation that answers both at once — because by the time a $6 million model inversion incident shows up in next year’s IBM report, the architecture decision that caused it will already be years old.
