Artificial intelligence is becoming more than a tool that answers questions. It can now help write emails, analyse documents, manage information and carry out tasks with limited human involvement. But as these systems become more capable, an important question arises: how much control should we give them?
Microsoft CEO Satya Nadella has raised concerns about AI safety. His message, shared publicly on October 10, 2026, is that advanced AI systems should be treated with the same caution that companies apply to potential security risks within their own organisations.
The concern is not necessarily that AI is deliberately trying to cause harm. It is that a powerful system, if given access to sensitive information and important operations, could make serious mistakes or be misused.
Why AI Should Not Be Trusted Blindly
Companies already take precautions against security threats, including unauthorised access to confidential files and misuse of employee accounts.
Nadella argues that advanced AI systems also need to be treated as potential sources of risk, particularly when they have permission to access company data or perform important tasks.
Consider an AI assistant that can read confidential emails, retrieve customer information and prepare financial reports. If it accesses the wrong information or carries out an incorrect instruction, the consequences could extend beyond a simple mistake in a chatbot response.
The lesson is straightforward: AI may be useful, but organisations should not automatically trust everything it produces or allow it unrestricted access.
Every AI System Needs an Emergency Stop
AI tools are increasingly capable of performing several tasks in sequence. A system might prepare an email, identify its recipients and send it. Another might update records or help manage business operations.
The greater the freedom given to these systems, the more important it becomes to have a way to intervene.
Nadella has highlighted the need for an emergency mechanism that allows an authorised person to pause or stop an AI system while it is working.
For example, if an AI assistant begins sending incorrect information to customers, a human supervisor should be able to stop the process before more messages go out.
This principle applies to any organisation using AI for important decisions or actions: there must be a clear way for people to intervene when something goes wrong.
AI Can Make Recommendations, but People Must Set the Rules
One of the biggest questions surrounding AI is who decides what it is allowed to do.
An AI system may be capable of analysing information and recommending a course of action. But that does not mean it should automatically have the authority to act on every recommendation.
Imagine an AI tool helping a company screen job applications. It could assist with organising applications and identifying relevant qualifications. However, the company must establish appropriate rules, check for unfair outcomes and retain human oversight of consequential decisions.
Similarly, an AI system helping a hospital organise appointments should operate within rules set by the hospital, rather than being allowed to make unrestricted changes on its own.
Being intelligent enough to suggest an action is not the same as having the authority to carry it out.
AI Must Leave a Record of What It Does
When a person makes an important business decision, an organisation may need to understand why it was made and who was responsible.
The same principle should apply when AI performs tasks.
If an AI system changes a database, prepares an important document or processes a business transaction, there should be a clear record of what happened. Where appropriate, that record should show the instructions given, the actions taken and the information used.
Such records help organisations investigate mistakes, identify misuse and establish accountability.
Without them, it can become difficult to determine whether a problem arose from incorrect information, a poorly designed system or an instruction that should never have been carried out.
The GAWAH Take: Human Responsibility Cannot Be Automated Away
Satya Nadella’s warning comes at a time when businesses are racing to introduce AI into everyday operations. The technology offers significant opportunities, from saving time on routine work to helping people analyse large amounts of information.
But efficiency should not come at the cost of accountability.
An organisation that allows AI to access confidential records, communicate with customers or make changes to important systems must think carefully about permissions, supervision and safeguards.
The same principle applies to individuals. Before allowing an AI tool to manage emails, access personal documents or perform actions on their behalf, users should understand what the tool can access and what it is authorised to do.
AI can make work faster and help people make better use of information. Yet the responsibility for setting boundaries, checking important outcomes and correcting mistakes remains with people and the organisations that deploy these systems.
The future of AI should not be about handing over control. It should be about using increasingly capable technology while keeping human judgement, oversight and responsibility firmly in place.
Editor’s note: The details attributed to Nadella should be checked against his original public post before publication.
