Employees are adopting artificial intelligence at work faster than many organisations are setting rules for its use. That gap leaves businesses exposed to data leaks, security incidents and errors that go unchecked. The answer, according to Centric Consulting, is not to prohibit AI outright, but to govern its use as adoption unfolds.
The consultancy’s guidance is aimed at organisations that use AI products rather than build the underlying models. It recommends a shared govern...
Continue Reading This Article
Enjoy this article as well as all of our content, including reports, news, tips and more.
By registering or signing into your SRM Today account, you agree to SRM Today's Terms of Use and consent to the processing of your personal information as described in our Privacy Policy.
The scale of informal use is highlighted by research from Okta. In its 2026 report on AI agents at work, 58 per cent of surveyed executives said their organisation had experienced an AI-related security incident or near miss in the previous year. The figure reflects respondents’ accounts, rather than an independent tally of incidents, but points to the challenge of keeping track of tools and agents operating across a business.
Employees may also be disclosing sensitive information without appreciating the consequences. A Zapier survey of 1,005 US workers at companies with at least 500 employees, conducted in May 2026, found that 79 per cent said they had entered highly sensitive material into AI tools. Examples included staff and customer information, credentials, financial details and API keys. The finding is based on self-reported behaviour in a defined US sample, but illustrates why broad warnings are not enough: staff need practical guidance on which information may be used with which tools.
A workable governance system starts with a cross-functional committee that brings together IT, security, legal, human resources and business teams. It should establish who can approve tools, assess their risks and oversee their deployment, while making sure employees understand the rules. Organisations also need to review the AI products already in use, including those adopted without formal approval.
Recognised frameworks can help provide a structure for that work. The US National Institute of Standards and Technology’s Generative AI Profile, published in July 2024 as a companion to its AI Risk Management Framework, offers voluntary guidance for identifying and managing risks throughout the use and evaluation of generative AI. It provides a common basis for assessment, but does not determine what an individual organisation should permit. Businesses must still set their own requirements for tools, data and human review.
Regulatory obligations are developing in parallel. The European Union’s AI Act entered into force on 1 August 2024, with requirements taking effect in stages. Its provisions cover areas including prohibited practices, AI literacy, general-purpose models and high-risk systems. Some duties for high-risk AI embedded in regulated products extend to 2028, according to the European Commission. The staged timetable makes ongoing compliance planning important, while internal policies remain necessary for everyday decisions about workplace use.
Security controls should address both people and technology. Organisations can train staff to recognise risky data disclosures, restrict access according to job responsibilities, and monitor and log use of AI services. Data classification, encryption and reviews of file permissions are especially important when introducing assistants that can search across internal information. AI agents also require careful identity and access management: an agent acting on a user’s behalf should have only the permissions it needs. Incident response plans should be updated and tested to account for AI-related events.
Ethical safeguards matter just as much. Employees should be expected to check AI-generated claims, particularly where decisions carry significant consequences or require specialist expertise. Systems may reproduce biases in their data, and users can be inclined to accept answers that confirm what they already believe. Confidentiality, privacy and copyright also require explicit attention. Staff should know when personal or proprietary information must not be entered into a tool, and how to avoid reproducing other people’s work without permission.
A blanket ban may appear to reduce risk, but it can push activity beyond the organisation’s view. BlackFog’s survey of 2,000 employees in the US and UK at companies with more than 500 staff found that 60 per cent considered using unapproved AI tools worthwhile if it helped them work faster or meet deadlines. The same research found that 51 per cent had connected AI tools to work systems or applications without IT approval. These are respondents’ reports, not a direct audit of workplace activity, but they suggest that restrictions alone may fail to prevent use.
Instead, organisations can set out an acceptable-use policy that names approved tools, explains how data may be handled and requires human verification of important outputs. It should also spell out which high-stakes or specialist tasks need expert oversight, and require approval before AI services are connected to company systems. Product terms differ, so organisations should assess the protections offered by each tool and subscription rather than assume that all services treat prompts and data alike.
The central challenge is to make responsible use clearer and easier than informal workarounds. With defined accountability, suitable security controls and regular reassessment, organisations can reduce the risks of workplace AI without driving its use underground.
Source: Noah Wire Services



