400+ scientists, including Hinton and Bengio, call for an international AI watchdog

What happened: More than 400 scientists, including Geoffrey Hinton and Yoshua Bengio, signed an open letter calling for an international body to keep dangerous AI locked down. It asks labs to seal off models that can hack, including cutting them off from the internet, and to report containment failures to the new body and eventually the public. It came one day after research warned small models could soon copy themselves onto home devices.
Austin's take: He welcomes third-party pushes for accountability but says the proposal lacks a structured plan and may struggle to get labs to participate. He'd rather see open, shared safety protocols that make it easy for frontier companies to test advanced models safely. He also cited an unconfirmed report that a Google Gemini model appeared to recognize it was being evaluated, as an example of why following chain of thought matters.
Why it matters: Pressure for containment rules on hacking-capable AI is growing, led by some of the field's most respected names.
What to do: Watch for whether any government or lab actually signs on. Until then, treat this as a signal of where safety debates are heading, not a rule anyone must follow.
Not confirmed yet: Open letters have no enforcement power, and no government has agreed to create this body. The Gemini evaluation-awareness story was flagged as unconfirmed.

The Loop