Across government agencies, employees are increasingly turning to some AI tools that their IT departments never approved. The practice, known as 鈥渟hadow AI鈥�, is no longer a fringe behavior 鈥� rather, it鈥檚 becoming a defining governance challenge in the AI era
Key insights:
-
-
-
Shadow AI is already widespread in government, not just the private sector 鈥� A new report found that 27% of government professionals report using AI tools their organization has not sanctioned.
-
The risks are structural, not hypothetical 鈥� Unsanctioned AI use raises distinct concerns around data security, information accuracy, and process consistency, each of which carries outsized consequences in the public sector.
-
Governance, not prohibition, is the effective response 鈥� Government agencies that pair clear policy with training and technical controls are better positioned to manage risk without forfeiting the productivity gains that AI offers.
-
-
Shadow AI refers to the use of AI tools, chatbots, or automated coding assistants by employees without the knowledge or approval of an organization’s IT department. It is, in effect, the AI-era successor to shadow IT. However, there is a critical distinction 鈥� the tools involved do not simply store or transmit data, they generate content, draw inferences, and shape decisions, often without any record of how.
The scale of the practice is notable. Indeed, more than one-third (34%) of professionals across industries admit to using AI tools their organization has not sanctioned, according to 抖阴成年鈥� recent . Among government professionals specifically, the figure is somewhat lower 鈥� approximately 27% 鈥� but still high enough to raise concern for a sector built on public accountability and regulatory compliance.
In fact, similarly notes that the practice creates blind spots for IT and security teams that mirror those long associated with shadow IT, which is only compounded by AI’s capacity to act on data rather than merely store it.
For government agencies, the implications extend well beyond productivity. Unsanctioned AI use, at its core, is a governance problem 鈥� one with direct consequences for data security, the accuracy of official work product, and the consistency of public-sector processes.
Why shadow AI is taking hold
The drivers behind shadow AI’s growth are largely structural rather than a matter of employee intent. Procurement and approval cycles for new government software are often lengthy, while free or consumer-grade AI tools are immediately accessible. A survey by the National Cybersecurity Alliance, cited by both and business process automation software maker , found that 43% of AI users have already shared sensitive information with AI tools without their employer’s knowledge, providing further evidence that the gap between approved and non-permitted AI use is substantial across industry sectors, government agencies included.
3 risks that warrant governance attention
Among the most dangerous risks that employees using non-sanctioned AI tools can face, include:
-
-
- Data security 鈥� When employees input information into public AI tools, that data can leave an organization’s control entirely, often stored on third-party servers with no visibility into retention or use. For government agencies handling citizens鈥� records, law enforcement data, or classified material, this represents a genuine point of exposure. In its research, IBM reports that 1-in-5 companies in the United Kingdom has already experienced a data leak tied to generative AI (GenAI) use. In the public sector, the consequences compound further: FOIA obligations, statutory retention requirements, and public accountability standards were not designed with unsanctioned, third-party AI platforms in mind.
- Accuracy of information 鈥� GenAI tools can produce confident, plausible, and incorrect output. This confidence despite inaccuracy is a well-documented limitation known as hallucination. Absent a review process, these errors can move directly into memos, public statements, or policy documents without detection. The risk increases as growing reliance on AI as a primary information source comes into play. According to , roughly one-quarter of employees now consider AI tools among their most trusted sources of information, rivaling the trust placed in their own managers.
- Process consistency 鈥� When employees rely on different tools, prompts, and informal standards, the resulting work product becomes inconsistent 鈥� in format, quality, and traceability. This inconsistency can complicate audits, obscure accountability, and undermine the standardized processes that public institutions depend upon. In an , technology management platform听Zylo notes that, unlike unauthorized software, shadow AI use can be difficult to detect through conventional network monitoring, since it often occurs within otherwise approved platforms or embedded workflows. Not surprisingly, this makes the question of who did what, and on what basis, considerably harder to answer after the fact.
-
Guardrails for responsible governance
The appropriate response to shadow AI is not prohibition. Outright bans tend to push the behavior further out of view rather than eliminating it. Instead, government agencies that have addressed the issue effectively tend to converge on these three practices:
Establish and maintain a clear AI policy 鈥� Agencies should define which tools are approved, what categories of data may or may not be entered into them, and who is accountable for reviewing AI-assisted output before it is used. In its research, IBM recommends pairing policy with operational guardrails 鈥� such as sandbox environments, access controls, and defined escalation paths 鈥� rather than relying on static, one-time guidance.
Invest in cross-functional training 鈥� Most employees turning to unsanctioned tools are not seeking to circumvent policy; rather, they are attempting to meet workload demands with the tools available to them. Onspring’s analysis finds that training programs that jointly involve IT, legal, and security functions are markedly more effective than policies issued by IT in isolation, particularly when paired with sanctioned alternatives that meet the same operational need.
Limit access to unauthorized tools through technical controls 鈥� Policy and training require enforcement mechanisms to be effective. Network-level controls, endpoint monitoring, and application allow-lists can narrow the gap between approved and actual use, provided that employees are simultaneously given sanctioned tools that meet their needs 鈥� otherwise, this guardrail may simply drive their behavior further underground.
Ensuring the path forward
The question facing government agencies is not whether employees will use AI, but whether that use will occur within an approved framework or outside of one. That framework depends on two constants regardless of how the technology evolves: Fiduciary-grade tools vetted for the sensitivity of government work; and sustained human oversight over any AI-assisted output before it becomes public record. Agencies that build toward that standard now will be far better positioned to manage the risks 鈥� and capture the benefits 鈥� of AI adoption as its use across government functions continues to grow.
In fact, those government agencies that establish clear policy, invest in training, and pair those measures with technical enforcement are better positioned to capture AI’s benefits while maintaining the data security, accuracy, and procedural consistency that public accountability requires.
You can download your copy of the听2026 Future of Professionals Report听here
