"The single biggest obstacle to AI adoption is distributed data," Kevin Bohan, Director of Product Marketing at Denodo, said — a concise diagnosis that frames the dilemma public agencies now face as they move agentic AI from pilots into mission-critical operations.
The AI trust gap: what it is and why it matters
Denodo frames an "AI trust gap" in the public sector as the disconnect between the requirements of trustworthy agentic AI and the capabilities of existing data architectures. According to the findings summarized in Denodo’s "The AI Trust Gap in the Public Sector" report, the problem is not just model maturity but the data foundation those models rely on: timely, explainable, and complete context for every case. The report warns that agencies that fail to build that foundation risk canceled projects and reduced public confidence; agencies that close the gap "early will be better positioned to act with confidence, speed, and accountability."
Real-time operational awareness and active context
Operational deployments demand live data. Seventy-three percent of organizations surveyed say AI data must be real-time or within the minute to be trustworthy — higher than the global average of 66 percent — reflecting needs in emergency response, fraud prevention, and safeguarding. Yet most public-sector data architectures were built for reporting: batch processing, pre-aggregated datasets and latency where AI needs immediacy. Denodo argues that agentic AI requires an "AI data layer" that provides active context — live, governed, semantically trusted data rather than stale snapshots or delayed copies.
Shared semantic context across distributed systems
Data access alone does not solve the problem. Sixty-one percent of organizations report difficulty identifying and preparing the most relevant and trustworthy data for AI. The report highlights inconsistent business meanings — where terms such as "citizen," "case," "risk," or "eligibility" vary across departments and systems — as a source of risk for incorrect recommendations or actions. Public sector initiatives typically touch a large number of sources: an average of 479 data sources per AI initiative, and more than one in five organizations access over 1,000 sources. Denodo emphasizes that consistent definitions plus governed, real-time access are both necessary to apply shared context inside operational AI workflows.
Governed action and runtime controls
Governance becomes determinative when AI moves from suggesting outputs to executing actions. The public sector reports the highest percentage — 72 percent — of respondents with difficulty in AI data security and access controls; 16 percent call these challenges "very difficult," nearly double the global average. The report stresses that governance must operate at runtime to enforce policies, permissions, and security controls before data is used by an AI application or agent. Auditability, explainability, and fairness are singled out as elements that help build public trust as agents take actions on behalf of agencies.
Cost, performance, and the economics of scale
Cost and performance optimization is a leading practical barrier: 40 percent of respondents identify it as the biggest data challenge for AI teams. Agentic AI intensifies pressures through repeated retrievals, multiple tool invocations, and extensive logging requirements. Denodo contends that a more efficient AI data layer — reducing unnecessary data movement, fragmented retrieval pipelines, and duplicated integration work — can lower token consumption and help scale AI more economically, yielding more safeguarding decisions per dollar and more responsive services within realistic budgets.
What this means for technologists, policymakers, and procurement leads
- Technologists and security teams will prioritize real-time access, semantic alignment, and runtime policy enforcement so agentic systems can act with explainability and auditability.
- Policymakers and regulators should expect to focus on enforceable governance at runtime — not just after-the-fact compliance — because auditability and policy enforcement are central to public trust.
- Public sector procurement and program leads will need to evaluate data-layer capabilities (live context, semantic trust, governed access) as a core criterion, since the report argues the future of AI initiatives depends more on trusted data foundations than on deploying additional models.
Denodo’s central message is plain: successful agentic AI depends on three capabilities — real-time operational access, consistent business context, and governance that supports accountability. "Ultimately, the future of AI is not about deploying more models, but about building a trusted data foundation." For agencies seeking to move from experimentation to scalable, trusted operations, that foundation is the operational challenge they must meet.
