Airlock EU-West (Dublin) operationalFreephone 1800 100 066
Back to CloudBlog

Nov 19, 2026 · CloudMount editorial

Prompt Data Leakage: The Copy-and-Paste Risk Irish Firms Overlook

Office buildings on Upper Mayor Street, Dublin Docklands
Cover: Upper Mayor Street, Dublin Docklands: William Murphy from Dublin, Ireland / Wikimedia Commons · CC BY-SA 2.0 · Cropped for layout.

The biggest generative AI risk for many organisations is not a hacker. It is a well-meaning employee pasting the wrong text into a chatbot.

A quiet, everyday risk

Generative AI tools are brilliant at tidying a report, drafting a reply or summarising a contract. To do that, they need the text, and staff will happily paste in whatever is in front of them: client emails, payroll figures, medical notes, even passwords sitting in a config file.

Once that information leaves your systems, you may lose control over where it is stored, how long it is kept and who can access it. Under GDPR, that can amount to a disclosure you are accountable for.

Why policies alone fall short

An acceptable use policy is essential, but people under deadline pressure do not reread policies before pressing enter. Training helps; so does giving staff an approved tool, so they are not tempted to use personal accounts.

The strongest control sits between the person and the model: checking each prompt for personal or confidential data and removing it before it is sent. That is the idea behind an AI data firewall such as CloudMount Airlock.

Practical takeaway

Pick one team and review a week of their AI use with them. Note what kind of information was shared. Agree which data must never be entered, provide an approved tool and consider automated screening for prompts. Small changes here prevent the most common, and most avoidable, AI data incidents.

General information only, not legal advice. Regulatory obligations depend on your role, context and the provisions in force.

Want to talk about your compliance?

Book a review