The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Shadow AI is AI used or deployed without the organisation’s governance—not just an employee pasting work into a public chatbot. The practical response is to find what is in use, understand the work it supports and the data or permissions it touches, then decide whether to approve, limit, replace or stop it. That makes shadow AI both a control problem and a signal about where employees need better tools.
What counts as shadow AI?
Microsoft Learn defines shadow AI as AI operating in an enterprise without governance. The defining feature is that the system falls outside the organisation’s controls, not whether it runs internally or externally. That makes the term broader than consumer chatbots.
As an Amazon Associate I earn from qualifying purchases.
It can include an employee entering work material into a public AI service; an unreviewed browser extension or task-specific AI tool; an internal agent that has no registered owner; or a developer connecting a model or API to business data or workflows without the relevant review. A tool can be familiar, useful and already in routine use while still being unmanaged.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitches“Unapproved AI” usually describes tools employees choose without organisational approval. “Unmanaged agent” describes an AI system that can access data or take actions without adequate ownership, oversight or controls. The categories can overlap, but they are not interchangeable: an unapproved chatbot may only respond to a prompt, while an internal agent may have permissions to act across connected systems.
Why do employees use unofficial AI?
Unofficial adoption often points to a mismatch between the work people need to do and the approved tools they can readily use. Employees may be trying to summarise documents, draft routine material, analyse information or automate repetitive tasks. A ban that leaves the underlying need unmet can drive use out of sight rather than resolve it.
In a Microsoft-commissioned Censuswide survey of 2,003 UK employees aged 18 and over, conducted in October 2025, 71% said they had used unapproved consumer AI tools at work. Among respondents who used shadow AI, 41% said they did so because the tools were familiar from personal life, and 28% said their company did not provide a work-approved option. These are UK employee survey findings, not estimates for every country, sector or workforce.
That evidence suggests a useful first question is not simply “Who broke the rule?” but “What were they trying to get done, and why was the approved route not meeting that need?” The answer helps distinguish a use that could be safely supported from one that needs tighter limits or must stop.
What can go wrong when AI is outside governance?
Data exposure and missing records
An unreviewed service may receive customer, employee or business information before anyone has established how it handles that data, who can access it or what retention arrangements apply. If the organisation does not know the service is in use, it may also lack the records needed to understand or respond to an incident. This does not mean every AI provider trains on submitted information; that must be checked for the specific service and its terms.
Rank #2
Unclear access and actions
An agent may have access to systems, files or credentials that have not been reviewed for its purpose. Without an owner, a defined scope and activity records, it can be difficult to determine what the agent can reach, what it has done or how to contain it. A connected agent therefore needs scrutiny beyond the question of which model it uses.
Security, compliance and unsuitable use
An unassessed service or integration can create security or compliance concerns, particularly when sensitive customer or financial data is involved. Whether a particular use violates a legal requirement depends on the facts and jurisdiction; the mere presence of shadow AI does not establish a specific breach or penalty.
The UK Information Commissioner’s Office discusses shadow agentic AI through future scenarios in which an employee sees a benefit but may not recognise wider risks or data-protection consequences. Those scenarios are useful prompts for risk planning, not incident counts or certain predictions.
What do the survey figures tell leaders—and what do they not?
Cloud Security Alliance survey announcements in 2026 point to visibility and control gaps around agents. Their results are not interchangeable: the surveys had different sponsors, dates and samples, and neither establishes prevalence for every organisation.
Rank #3
| Finding | Survey context | How to interpret it |
|---|---|---|
| 82% reported unknown AI agents in their IT infrastructure; 65% reported AI-agent-related incidents in the prior 12 months. | Cloud Security Alliance survey commissioned by Token Security, conducted online in January 2026 with 418 IT and security professional responses. | Among respondents reporting incidents, 61% reported data exposure, 43% operational disruption and 35% financial losses. These are respondent reports, not an independently verified incident rate for all organisations. |
| 54% reported between 1 and 100 unsanctioned AI agents; 31% reported formally adopted AI-agent-use policies. | Cloud Security Alliance survey commissioned by Zenity, announced April 16, 2026, with 445 IT and security professional responses gathered in September and November 2025. | This is a separate survey from the Token Security-commissioned survey and should not be combined with its results. |
Together, the surveys make a case for asking whether an organisation can account for agents and their actions. They do not show that any single product or policy will eliminate risk.
How should an organisation turn discovery into governance?
A workable programme connects visibility to decisions and employee needs. A list of tools is a start, not the result: leaders need to know what each system does, what it touches and who is accountable for it.
1. Discover tools, agents and integrations
Look beyond a list of approved software. Include employee-facing services and browser extensions, internal agents, developer workflows, APIs connected to business data, and SaaS products with AI features. Review relevant identity, endpoint, network, procurement and application records where available, and provide a clear route for staff to disclose tools they have adopted. Microsoft Learn recommends enterprise-wide visibility across agents, users, devices and external AI services.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Record what is known and mark uncertainties for follow-up. A conventional application inventory can miss an agent embedded in a workflow or a model call made through an integration.
2. Understand the task and the data flow
For each discovered use, ask what task it supports, who uses it, what information is entered or retrieved, which systems are connected, and whether the AI only generates suggestions or can take actions. Ask the people doing the work as well as the technical owner: observed activity explains what is happening, while employee context explains why.
3. Assign an accountable owner and assess risk
Record a business owner and a technical contact. Assess data sensitivity, access permissions, intended purpose, user population, affected systems and possible impacts. For an agent, document its allowed actions and the boundaries around them; for an external service, establish how information is handled and what controls apply.
NIST’s Generative AI Profile is a voluntary, cross-sector companion to AI RMF 1.0. It can help organisations incorporate trustworthiness considerations into AI design, development, use and evaluation. It is a risk-management reference, not an inventory system that discovers deployed tools for you.
4. Match controls to the use
Choose an outcome for each use: approve it as-is, approve it with conditions, investigate further, restrict data or actions, or block it if the risks cannot be made acceptable. Controls for agents may include narrow permissions, defined operating boundaries, activity monitoring and human authorisation for consequential actions. Plan how to revoke credentials and retire the agent when it is no longer needed.
Do not assume that a human approval step makes every action safe or that monitoring alone prevents harm. Set controls to the actual data, permissions and consequences involved, and decide who reviews exceptions and incidents.
5. Provide an approved route that solves the same problem
Where a useful activity can be supported safely, make the approved option easy to find and explain how employees can access it. Include practical guidance on what data may be used, how to check outputs and how to raise questions. If the sanctioned tool is difficult to access or does not support the task, revisit the offer rather than relying only on reminders to comply.
The UK Cabinet Office’s human-centred “People Factor” and “Mitigating Hidden AI Risks Toolkit” address engagement, training, support, risk management and monitoring. They offer an approach to involving people in adoption rather than treating governance as a technical-only exercise.
6. Review the inventory and controls over time
AI services, integrations and agent permissions can change. Revisit owners, data flows, purposes, access, monitoring and incidents periodically, and when a material change occurs. Check whether the approved alternative still meets the work need. Include credential removal and decommissioning in the lifecycle, not just initial approval.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Should you start with a framework or discovery software?
These address different parts of the problem. A framework helps structure decisions; discovery or enforcement software may help identify activity or apply controls. Neither substitutes for ownership, decisions about acceptable use or a useful approved offer.
| Approach | Useful for | What it does not establish by itself | Choose it when |
|---|---|---|---|
| Governance framework or process, such as NIST’s Generative AI Profile | Consistent risk assessment, roles, review criteria and lifecycle decisions. | Which tools, agents or integrations are actually deployed in your environment. | You need a repeatable way to evaluate and govern known uses. |
| Discovery or enforcement software | Potentially identifying services or agents, surfacing activity, monitoring or enforcing selected policies, depending on the product and configuration. | Whether a use is appropriate, who should own it, or whether the product covers every relevant system and workflow. | You need technical visibility or control that existing processes cannot provide, after defining the scope and requirements. |
When evaluating software, ask whether it covers SaaS, endpoints, internal agents and APIs; whether it can help establish owner and purpose; what it reveals about data and permissions; and how it supports risk classification, audit, monitoring, enforcement and decommissioning. Also assess integration with existing controls, usability for employees, and the operational burden of maintaining it. The cited sources do not provide an independent vendor comparison, so vendor capability claims should not be treated as comparative proof.
What does success look like?
Success is not a claim that no employee will ever try an unapproved tool. It is the ability to account for meaningful AI use, understand the work and information involved, assign responsibility, make proportionate decisions and respond when something changes. Just as importantly, the organisation can see where its approved offer is failing and improve it before useful work moves further outside governance.
Recommended Free Tools
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
