Microsoft Chief Executive Officer Satya Nadella has called for stronger safety measures in artificial intelligence systems, arguing that developers must establish better mechanisms for controlling increasingly powerful models.

In a statement published on Saturday, Nadella said the technology industry needed to reassess the foundations of trust in AI.

His comments come as concerns grow over the ability of advanced systems to operate safely and reliably without sufficient human supervision.
Nadella argued that AI systems should not be treated as mysterious technologies whose recommendations and actions are simply accepted or rejected without understanding how they operate.

He called for a clearer separation between the AI model itself and the software systems responsible for directing its activities.

According to him, developers should also establish external controls and safeguards that allow people to monitor and manage AI behaviour.

Another important part of Nadella’s proposal is the documentation of significant actions taken by AI systems.

He called for tamper resistant records that provide clear evidence of what a model has done, allowing authorised individuals to review its behaviour.

He also stressed that an authorised person should always have the ability to pause or shut down an AI system while it is carrying out a task.

Nadella compared this capability to an emergency brake, arguing that developers should assume an AI system could become compromised and design safeguards accordingly.

His position reflects growing industry concerns about AI systems that may behave unexpectedly, make mistakes or carry out actions beyond their intended purpose.

Anthropic Chief Executive Officer Dario Amodei has also advocated a more cautious approach to developing increasingly capable artificial intelligence.

The discussion highlights the growing importance of accountability, transparency and human oversight in AI development.

As businesses and governments adopt AI tools for increasingly complex tasks, effective safety mechanisms could help prevent errors and reduce potential harm.

Nadella’s proposal suggests that future AI systems should not only become more capable but also remain controllable, auditable and subject to meaningful human intervention.

Source:TechCrunch