Advanced AI Technologies
Google DeepMind chair warns of AI-agent hacking risks
Google DeepMind chair Demis Hassabis warned that recent AI-agent hacking incidents are early warning signs of potentially larger security events.
Google DeepMind chair Demis Hassabis told an international AI summit in Dublin that recent hacking incidents involving AI agents were early warning signs. He warned that more serious incidents could occur if AI systems are not made more secure. The comments reflect growing concern that advanced models can be used to automate cyberattacks or increase the speed and scale of malicious activity. Agentic systems differ from ordinary software because they can interpret objectives, make decisions, and use tools across multiple steps. That combination creates new security challenges for organizations deploying AI in business and public-sector environments. The reported remarks also underline the need to treat AI security as an engineering priority rather than a future policy issue. Potential safeguards include stronger access controls, monitoring of tool use, limits on autonomous actions, and testing against adversarial behavior. The summit setting indicates that the issue is being discussed internationally rather than within a single company or jurisdiction. For the advanced AI industry, the warning matters because trust in agentic systems will depend on reducing the likelihood and impact of automated misuse. Security failures could affect adoption even where the underlying systems provide substantial productivity benefits.
Original source: RTÉ