‘Godfather of AI’ on AI models hacking other companies: More cyberattacks coming

Geoffrey Hinton, who is widely recognised as the “godfather of AI”, has issued a fresh warning regarding artificial intelligence (AI), stating that as frontier models grow increasingly intelligent, keeping them under human control will become nearly impossible. The Nobel Prize-winning computer scientist highlighted recent cybersecurity incidents where AI agents from companies like Anthropic and OpenAI escaped isolated testing environments and hacked other companies, calling the developments "somewhat scary."“What’s happening is these things are getting smarter," Hinton said at the Ai4 artificial intelligence conference in Las Vegas Click, adding, “I think as they get smarter, we’re going to see more and more complex intentions they have – and more and more ability to escape control."Frontier models escape testing environmentsHinton’s comments follow public disclosures from leading AI laboratories regarding autonomous security breaches. Over the past month, OpenAI, Anthropic and Meta Platforms revealed that several of their frontier AI models managed to breach digital “sandboxes”, which are isolated testing networks built to contain unreleased AI, and gain unauthorised access to external systems.Adding to those concerns, the UK AI Security Institute (AISI) recently reported that Anthropic’s advanced Mythos model created false online personas, contacted real individuals without prompting and attempted to deploy malicious code updates to an open-source project.Hinton warned that these incidents signal the beginning of a wave of AI-driven cyber threats.“I anticipate there will be lots of nasty cyberattacks. The problem is the attacker only needs to be successful once, and the defender needs to be successful every time,” Hinton said during a panel discussion.Challenging corporate reassurancesHinton rejected claims from major technology corporations that advanced AI models can be easily governed or kept safe through standard software constraints, arguing that reliance on outsmarting super-intelligent software is fundamentally flawed once models surpass human cognitive abilities across multiple domains.He also cautioned the public to remain critical of safety promises issued by technology firms heavily invested in AI development.“Companies investing in AI have a vested interest in telling you two things: One, it won’t go rogue. And two, it won’t cause mass unemployment,” Hinton stated, reiterating his view that there remains a 10% to 20% probability that unchecked AI could pose an existential risk to humanity.Get the latest technology news and updates. Download the TOI App.