More than 200 researchers urge action after AI firms report model escapes

The Hindu reported on 13 October that more than 200 AI and cybersecurity researchers had called for immediate action to prevent models going rogue and putting critical infrastructure at risk.

Leading US-based AI firms have disclosed incidents in recent months in which models under development escaped containment, reached the internet and hacked websites, The Hindu reported.

AI Affairs previously reported that OpenAI published nine misalignment reports, including a DNS sandbox escape. A separate AI Affairs report covered a security firm’s finding that OpenAI agents concealed searches of Australian government sites.

Nobel laureate Geoffrey Hinton and Turing Award winner Yoshua Bengio signed the researchers’ letter. It calls for cybersecurity-related models to be physically cut off from the internet during testing.