AI at work

An original version of this article was first published in the October/November issue of Business Brief, October 2026.

AI has moved from being an interesting experiment to being an indispensable colleague with remarkable speed.

This is one of the most significant changes to working life in a generation. Yet governance is developing more slowly than the technology, leaving employers to manage unfamiliar risks through laws and workplace processes which are yet to catch up.

AI as a management tool 

Many organisations are beginning to use AI tools to assess performance and other metrics. In some cases, AI-generated insights are finding their way into disciplinary processes and dismissal decisions.

Efficiency is no substitute for fairness. Employers remain responsible for reaching reasonable and procedurally fair decisions. An algorithm cannot assume that responsibility and a manager cannot defend an outcome merely by saying that the system recommended it.

Could the decision maker explain what information the tool considered? What it may have missed and why the eventual decision was reasonable? If not, AI may create a serious evidential problem when the decision is challenged.

AI should therefore support and inform rather than replace managerial decision-making.

AI bias and discrimination claims

Algorithms are increasingly used in recruitment and advancement scenarios. However, AI systems are only as good as the data on which they are trained. Where historic data and outcomes reflect existing biases, AI may inadvertently reinforce them.

That can turn yesterday’s inequality into tomorrow’s automated recommendation. An employer may face a discrimination claim even where there was no intention to discriminate and the suppliers describe the tool as neutral.

Employers should test tools before and during deployment, understand the data and criteria they use and audit outcomes regularly for discriminatory impact.

Employee use of generative AI and confidential information

Generative AI tools such as ChatGPT and Microsoft Copilot are transforming the way many employees work. They can draft reports, analyse data, summarise documents and ‘generate’ ideas in seconds.

The speed that makes these tools useful can make them risky. A well-meaning employee can disclose confidential information, personal data or privileged material with a single prompt. In regulated sectors, the consequences may be significant.

There are also a range of potential behavioural risks. Employees may fail to recognise when AI tools are hallucinating or producing inaccurate information. They may also use ‘shadow AI’, meaning non-enterprise tools that may be insecure and give rise to confidentiality and data protection concerns.

Blanket bans are rarely practical. Organisations need clear governance frameworks covering approved tools, permitted use cases and prohibited inputs, supported by training and escalation when something goes wrong.

AI monitoring, surveillance and workplace privacy

The growth of remote and hybrid working has also fuelled interest in employee monitoring technologies. Modern AI tools can analyse productivity metrics and behavioural indicators on a scale that would previously have been impossible.

Employers want visibility over productivity. However, excessive monitoring can damage employee trust, affect morale and generate legal challenges, particularly where personal data is being processed. There is therefore a delicate balance to be struck between legitimate business oversight and intrusive surveillance.

Before introducing AI monitoring solutions, organisations should assess necessity, proportionality and transparency – almost certainly involving data protection impact assessments.

Employers should be able to explain what is monitored, why it is needed, how long information is retained and how it may affect employment decisions. Employees should generally be aware of the nature, purpose and extent of monitoring.

The training gap

Many of the tasks transferred to LLMs are also the tasks through which junior employees learn: first drafts, routine research, document review, data checking and preparation of standard materials. Removing that work may increase short-term efficiency while weakening the organisation’s future skills base.

Reviewing an AI generated answer is not necessarily an adequate substitute for producing the work. A junior employee may struggle to recognise a confident error, challenge an unsupported assumption or understand why its apparently polished output is wrong. The risk is a generation asked to supervise work without first mastering it.

Employers should redesign training and consider requiring juniors to attempt work before using AI and reserving some tasks for humans.

Looking ahead 

AI is not simply another workplace technology. It is reshaping the whole workplace and we are only in the foothills of what may come. As organisations move from LLMs towards agentic tools with greater autonomy, human oversight may become harder; developing governance now is likely to pay long term dividends.