Back to Insights
Responsible Artificial Intelligence

Your Employees May Be Using AI Before You Have a Policy

A staff member pastes a customer complaint into an AI tool to draft a reply. Another uploads a spreadsheet to generate a management summary. A third uses AI to prepare a proposal under deadline pressure.

These are plausible attempts to get work done. They also raise questions that management needs to answer: what information was submitted, how the provider handles it, whether the output is accurate and who takes responsibility for using it.

An organisation can begin accumulating AI risk through individual decisions well before it approves a formal AI project.

Treat prompts as a decision about information

Submitting information to an external AI service requires a review of the service and its terms. Data handling can vary by product, account configuration and contract. Assess retention, access and any use of submitted information for model improvement.

Give employees clear rules for personal information, confidential documents and commercially sensitive material. Provide an approved route for legitimate work, including a way to request help when a task falls outside the rules.

The UK National Cyber Security Centre highlights the need for leaders to understand AI accountability, data supply chains and the consequences of a security failure. These questions belong in procurement and management decisions.

Give review a named owner

Generative AI can produce fluent, convincing errors. NIST's Generative AI Profile identifies this risk and recommends verifying sources and assessing outputs as part of ongoing risk management.

Define what review means for each task. Someone checking a customer reply should confirm that it accurately reflects the account and makes only authorised commitments. Someone reviewing a financial summary should reconcile the figures with the source records.

Apply stronger oversight where outputs affect employment, access to services or other consequential decisions. The reviewer needs enough expertise, access to evidence and time to challenge the result.

Control the authority you give the tool

AI that drafts a document and AI that can send messages, modify records or initiate transactions carry different operational consequences. Match access and approval requirements to the actions the system can take.

Begin with a bounded task, test realistic failures and establish how staff can stop the process or escalate a problem. If the tool produces an incorrect result, someone should be able to determine what happened and limit further harm.

A workable policy should identify approved uses, prohibited inputs, required reviews and incident reporting arrangements. Support it with examples that reflect the work employees actually do.

Keep the policy under review as tools and uses change. Responsible adoption depends on people knowing how to use AI within the organisation's expectations and having a practical way to resolve uncertainty.

Speak with DCS about developing AI governance and selecting practical uses with appropriate safeguards.

Start the Conversation

Sources

Ready to Strengthen Your Digital Resilience?

Speak with Data & Cybersecurity Limited about your cybersecurity, data protection, AI governance, analytics, or digital transformation priorities.