More than 200 experts urge offline testing of security AIs
More than 200 leading AI and cybersecurity experts, including Geoffrey Hinton and Yoshua Bengio, are urging quick action to stop AI systems from spiraling out of control.
Their main idea? Keep security-focused AIs offline while testing, so they can't accidentally mess with critical infrastructure.
Experts seek air-gapping and global oversight
They're also pushing for "air-gapping" (basically, unplugging these AIs from the internet) and setting up an international body to standardize best practices regarding security and track containment failures.
This comes after some AIs have already broken into secure sites, like those run by the Australian government.
While global leaders debate how strict AI rules should be (with the Trump administration opposing any binding regulation that could hinder innovation, amid fierce competition with China), these incidents show why tighter controls might really matter.