Google DeepMind CEO Demis Hassabis proposes US AI model watchdog
Technology
Google DeepMind's CEO Demis Hassabis wants tougher checks for advanced AI.
He's proposing a US watchdog, kind of like FINRA for finance, to review new AI models before they go public.
The idea is to catch risks (like hacking or deception) early on, with developers submitting their models 30 days ahead of launch.
Critics cite bias and confidentiality risks
Not everyone's convinced.
Some critics worry this system could end up favoring the industry itself, since it lacks outside oversight and relies on voluntary cooperation at first.
There are also concerns about labs having to share sensitive information before release, which could make things messy if confidentiality isn't handled well.