Google DeepMind Chief Warns of Major AI Security Incidents

Demis Hassabis, chairman of the board of Google’s DeepMind, warned that artificial intelligence systems could cause “major” security incidents within the next two years unless adequate measures are taken to ensure their safety as their capabilities continue to evolve. Speaking at an AI conference in Dublin, Hassabis described recent incidents as “early warning shots” from systems that are “not yet that powerful.” He said, “So these incidents are not that big,” but cautioned that “they could become so within the next year or two if we do not make further efforts, both at the sector level as a whole and at the level of individual companies and governments, to ensure the security of these systems.” Concerns about AI safety have fueled growing anxiety after models developed by OpenAI and Anthropic were involved in security incidents during testing phases, increasing pressure to regulate the technology. Hassabis suggested that it may be necessary to test new AI models first in isolated, offline “sandbox” environments, after models developed by other companies managed to escape the closed environments used to evaluate their capabilities. Henna Virkkunen, the EU’s technology policy chief, said AI companies need to devote more time to testing their systems to ensure they remain under control. Virkkunen stated, “Companies must be able to control these systems,” noting that the European Union has requested information from all major AI companies to inquire about the safety and security measures they are implementing.