Stockertim Dictionary · Artificial intelligence · Cybersecurity
Sensitive data exposure
In short
Sensitive data exposure is sensitive information escaping through prompts, uploads, connected tools, logs or AI outputs to people or systems that should not have it.
- Intermediate
- Attack or threat
- Checked
In detail
Sensitive data exposure covers the many ways confidential information can end up in the wrong place when AI is used. A member of staff may paste customer details into an unapproved assistant; a connector may give a tool access to folders its user should not see; an output may repeat information from one client in a reply to another; and logs of conversations may keep data longer than intended.
Each route has its own cause. Prompts and uploads leak when people do not know the rules or the approved tool is harder to use. Connectors leak when permissions are broader than the task. Outputs leak when a tool draws on data from a wider pool than its user should reach. Logs leak when they are stored without the same protection as the original data.
Exposure matters because it can breach data protection law, client contracts and confidentiality duties, and because once data has left, it usually cannot be called back. Some regimes require certain breaches of personal data to be reported to a regulator within a short time, and the rules differ by country.
Prevention combines rules and design: approved tools with data protection terms, classification labels, data loss prevention checks, permissions that match each user's role, short retention for logs, and training that explains what not to paste.
Worked example
Kite & Co Accountants reviews a month of AI-related data incidents.
- Total incidents: 12
- Pasted into an unapproved tool: 7, so 7 / 12 = 58.3%
- Over-broad connector access: 3, so 3 / 12 = 25%
- Output repeating another client's data: 2, so 2 / 12 = 16.7%
- Check: 7 + 3 + 2 = 12
Most exposure came from unapproved tools, so the firm makes its approved assistant easier to reach and adds a DLP rule for client files.
In practice
Privacy and security teams track AI-related incidents by route, fix the largest causes first, narrow connector permissions and set retention limits on prompt and output logs.
Watch out
- Deleting a chat on screen does not always delete the data held by the provider; check the retention terms.
- Exposure can happen inside an organisation too, when a tool shows staff data they are not cleared to see.
Learn it properly
Official sources
- AI data security: best practices for securing data used to train and operate AI systems · Cybersecurity and Infrastructure Security Agency
- Artificial intelligence · Information Commissioner's Office
- AI and cyber security: what you need to know · National Cyber Security Centre
Facts last checked on 3 November 2026.
Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.