US AI Safety Rules Might Miss Some Big Players

Big news from the United States this week about how they plan to keep artificial intelligence safe. The US government is working on a framework – think of it as a set of guidelines – to test AI for safety. However, it looks like this testing won't apply to all major AI models.

Specifically, the rules might exempt what are called 'open-weight' AI models. These are models like those from Meta (the company behind Facebook) and Nvidia, where much of the underlying technology is openly shared. In contrast, 'closed' AI systems from giants like Google and OpenAI, which keep their technology under wraps, will be included in the safety tests.

This decision has sparked a bit of a debate. Some argue that open-weight models, because their components are more accessible, can be more easily scrutinised and improved by a wider community of experts. Others worry that if these models aren't formally tested, potential risks might be overlooked, especially as AI becomes more powerful and widespread.

For everyday Australians, especially those running small businesses or just using AI tools, this is worth keeping an eye on. While these are US rules, they often set a precedent for how other countries, including Australia, might approach AI regulation. The core question is always: how do we make sure this powerful new technology is safe and works for us, not against us, regardless of who built it?