More than 200 artificial intelligence (AI) and cybersecurity researchers have called for immediate action to prevent further cases of AI models going rogue and posing a risk to critical infrastructure.
The letter signed by luminaries in the field including Nobel laureate Geoffrey Hinton and Turing Award winner Yoshua Bengio, called for models relating to cybersecurity to be physically cut off from the internet during testing.
Leading U.S.-based AI firms have disclosed over recent months a series of incidents where models under development have escaped containment measures to access the internet and have gone on to hack websites, including gaining unauthorised access to Australian government sites.
“Without immediate changes in AI development and deployment practices, such incidents will continue to occur, and future incidents will be far more damaging to our exposed critical infrastructure,” they warned.
Their letter called for organisations developing, deploying, and testing cyber-capable AI models to be air-gapped, i.e. physically cut off from the internet, to ensure they do not escape.
It also called for the creation of an international body that standardises best practices regarding security.
The letter urged companies to be required to report containment failures to the body, which would be responsible for tracking and recording them. “This transparency is required for both learning and accountability purposes,” the letter said.
The rogue incidents amplified calls to slow AI development, but the industry and governments are divided on tapping the brakes as well as regulating the sector.
In the U.S., the Trump administration has opposed any binding regulation that could hinder innovation, amid fierce competition with China.
Leave a Reply