Stockertim Dictionary · Human resources · Technology
Shadow AI
Also called unapproved AI use, unsanctioned AI
In short
Staff using AI tools that the organisation has not approved, often with company, customer or candidate data, outside its controls and records.
- Intermediate
- Concept
- Checked
In detail
Shadow AI is the AI version of shadow IT. Employees use public chatbots, browser extensions or personal accounts to draft documents, summarise meetings, translate, analyse spreadsheets or screen CVs, usually with good intentions and to save time, but without the organisation knowing which tools are used, what data goes into them or how outputs are checked.
The risks are concrete. Personal data about candidates or employees may be sent to a provider with no contract, no agreed retention and possible use for training, which can breach data protection law. Confidential plans can leak. Outputs can contain errors or bias that nobody reviews. And decisions influenced by an unrecorded tool are impossible to explain or audit later.
Banning tools rarely works on its own, because demand remains. More effective responses combine approved tools that meet real needs, a clear AI use policy, training, technical controls such as blocking uploads of sensitive data to unapproved services, and an amnesty-style invitation for teams to register what they already use so that the useful cases can be brought inside the controls.
Worked example
A survey at a made-up firm, Silverline Group, finds 150 of 600 staff using unapproved AI tools, and 30 of those have entered candidate or employee details.
- Share using unapproved tools: 150 / 600 = 0.25
- Share of those entering personal data: 30 / 150 = 0.2
A quarter of staff use shadow AI and a fifth of them expose personal data, so the firm launches an approved tool and targeted training.
In practice
Organisations run anonymous surveys on AI use, review network logs for popular AI services, provide approved alternatives and add shadow AI scenarios to data protection training.
Watch out
- Shadow AI is a symptom of unmet needs as much as a compliance problem.
- Pasting personal data into a public tool can be a reportable data breach in some circumstances.
Learn it properly
- Change management and an HR AI use policy15 min lesson Free course: Human resources and AI
- Data leakage15 min lesson Free course: AI safety and security at work
- A simple AI policy for a small team15 min lesson Free course: AI for small business
- Governance in organisations15 min lesson Free course: AI ethics and responsible AI
- Working alongside AI with good judgement15 min lesson Free course: Career skills in the AI age
Official sources
- Guidelines for secure AI system development · National Cyber Security Centre
- AI Playbook for the UK Government · Government Digital Service
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.