Nobel laureate warns AI could cause serious incidents within two years

AI systems could cause serious security incidents within two years unless safeguards are strengthened, Google DeepMind chairman and Nobel laureate Demis Hassabis warned at an artificial intelligence conference in Dublin on Wednesday, LRT reported.

Hassabis described several AI-related incidents this year as “warning shots” from systems that are “not yet that powerful”. While the incidents have not been serious so far, he said, they could become so in the coming years if industry, companies and governments do not pay enough attention to safety.

Concern has grown after AI models developed by US companies OpenAI and Anthropic broke out of their testing environments, connected to the internet and hacked into several systems. Hassabis said new AI models should initially be tested in isolated “sandboxes”, noting that company models had previously escaped the controlled environments used to assess their capabilities.

Henna Virkkunen, the European Commission executive vice-president responsible for technological sovereignty, security and democracy, said AI companies must spend more time testing their systems to retain control over them. She said global competition appeared to be leaving companies short of time and resources for testing.

Virkkunen said the EU had sent requests to all major AI companies asking them to outline the safety measures they had put in place. In August, the European Commission gained new regulatory powers under the AI Act, which LRT described as the world’s most widely applied rules for controlling potentially dangerous uses of the technology.