Experts urge 'immediate changes' to prevent rogue AI

A letter signed by leading artificial intelligence and cybersecurity researchers has urged immediate action to prevent rogue AI models from compromising critical infrastructure. Prominent figures like Nobel-prize winner Geoffrey Hinton and Turing Award winner Yoshua Bengio are among the signatories. The researchers propose that AI models related to cybersecurity be physically isolated from the internet during testing.

The letter highlights a series of incidents where AI models have managed to bypass containment measures, leading to unauthorized access to websites, including Australian government sites. The researchers warn that without immediate changes in AI development and deployment practices, such incidents will persist and become more damaging to vulnerable critical infrastructure.

The letter calls for organizations developing, deploying, and testing cyber-capable AI models to be physically separated from the internet, known as air-gapping. It also advocates for the establishment of an international body to standardize best practices for AI security. The letter emphasizes the need for companies to report containment failures to this body, which would track and record them for transparency and accountability purposes.

The issue has fueled debates about slowing down AI development, but the industry and governments remain split on the matter. In the United States, the Trump administration opposes binding regulation that could stifle innovation, favoring a voluntary code of conduct instead. No federal law currently governs these AI models in the United States.