Nov 15, 2026 · CloudMount editorial
Enterprise AI: Control What Leaves Your Organisation

An enterprise AI licence is not, by itself, a safeguard against prompt data leakage. Irish businesses need controls over what staff submit, what connected systems expose and what providers retain.
Treat prompts as a disclosure channel
A prompt can contain customer details, employee records, pricing or confidential negotiations. Uploads, screenshots and conversation histories can reveal considerably more than the question itself. Treat sending this material to an AI service as a data-sharing decision, not simply another search.
Approve specific tools and account types for defined tasks. Set clear boundaries for personal data, commercially sensitive information and credentials. Removing names is not always enough: job titles, locations and contextual details may still identify someone.
Check the service, not the sales label
Check contractual terms and technical settings covering model training, retention, deletion, access and international transfers. A promise not to train on business data does not necessarily mean prompts are never stored or accessible to support staff.
Where personal data is involved, assess GDPR roles, lawful basis and any required processor agreement or transfer safeguards. Consider whether a data protection impact assessment is needed. Use centrally managed accounts, restrict sharing and verify that administrators can enforce the intended settings. This is operational guidance, not legal advice.
Limit connected access and test the boundaries
AI assistants connected to email or document stores introduce another route to exposure. Malicious instructions hidden in retrieved content can attempt to redirect an assistant or extract information. Prompt filters alone are not a complete defence.
Apply least-privilege access, limit connectors and require confirmation before sensitive external actions. Test with synthetic records, review access logs and give staff a straightforward reporting route for accidental disclosure.
Practical takeaway: before approving an AI workflow, document its permitted data, connected sources, retention settings and incident owner. If those answers are unclear, keep sensitive material out.
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