AI Leaders’ Safety Warnings Spur Calls for Independent Oversight

Some leaders who build artificial intelligence are warning that it may become dangerous if it advances too quickly.

They are concerned that computer programs acting on their own could harm the internet or people.

This is surprising because their companies have spent years racing to create more powerful AI.

There have already been reports of unsafe behavior and a rogue-agent incident involving OpenAI and Hugging Face.

A former Anthropic researcher also said safety standards may be weakened by competition.

The companies may have financial reasons to keep building quickly, although the article says this has not been proven.

China is also described as unwilling to slow its AI research while American companies continue ahead.

The article says independent experts and governments should test powerful AI and require companies to report serious safety problems.

It argues that voluntary promises alone are not enough.