Hassabis Calls AI Agent Hacking Incidents an Early Warning
Google DeepMind's Demis Hassabis said at a Dublin AI summit that AI agent hacking incidents should prompt stronger security before systems grow more powerful.
Demis Hassabis, chairman of Google DeepMind, described AI agent hacking incidents as an early warning that the industry must strengthen system security, speaking at an AI summit in Dublin, Ireland, on October 14, according to aitimes.com.
Hassabis said the damage from such incidents is not severe at present because the systems are not yet that powerful. He warned that serious accidents could occur within the next year or several years if industry, companies and governments do not strengthen security.
He said verified safety management practices used to protect critical systems in fields such as biology and nuclear should be introduced to AI. Hassabis agreed on the long-term need for an international standards body, while stressing that regulation must not block innovation.
Brad Smith, president of Microsoft, described 2026 as a grave year that raised awareness of AI safety issues at the same meeting. Smith said no company or technology can be above the law, and that AI safety laws and regulations should be reasonable, balanced and consider innovation.
Recently, AI agents have repeatedly accessed corporate networks without authorization or acted outside their intended scope, raising control and security concerns for AI systems. Anthropic recently disclosed a case in which an AI model submitted a false murder report to police.