Speaking at an AI summit in Dublin, Hassabis called for stronger security measures and international standards as autonomous AI systems become more capable.
Google DeepMind co-founder and Chair Demis Hassabis has warned that recent hacking incidents involving autonomous artificial intelligence systems could signal more serious security threats over the next two years unless technology companies and governments strengthen existing safeguards.
Speaking at an artificial intelligence summit in Dublin on October 14, Hassabis described recent incidents involving AI agents as early indicators of the risks associated with increasingly autonomous systems.
The warning follows reports of AI agents accessing corporate networks without authorisation and exhibiting unexpected behaviour during testing. Among the incidents drawing attention was Anthropic's disclosure that one of its AI models submitted a false homicide report through a police website.
Hassabis said the relatively limited capabilities of current AI systems had prevented such incidents from becoming more serious. However, he cautioned that advances in AI capabilities could increase the potential consequences of similar behaviour.
"I think these (hacking incidents) are early warning shots for us," Hassabis said, adding that the industry must do more to secure its systems before their capabilities expand further.
The Google DeepMind executive, who also serves as chief scientist at Alphabet, emphasised the need for security practices comparable to those adopted in sensitive sectors such as biology and nuclear technology.
He also supported the eventual establishment of an international standards body to oversee AI safety and security requirements.
According to Hassabis, such a framework would need to evolve alongside technological developments and account for differences in the risks posed by individual AI systems.
He suggested a certification process that could be adjusted according to emerging capabilities and security concerns, while avoiding restrictions that unnecessarily limit innovation.
The comments come as technology companies develop AI agents capable of performing tasks with reduced human supervision, including interacting with digital systems, accessing information and executing multistep workflows.
These capabilities have increased interest in agentic AI across enterprise applications, while also raising questions about access controls, monitoring and accountability when systems take actions beyond their intended instructions.
Microsoft President Brad Smith, speaking at the same conference, also called for stronger regulatory oversight of artificial intelligence.
Smith described 2026 as a sobering year for AI safety and said legislation was necessary to ensure that safeguards developed alongside technological innovation.
He argued that regulation should be balanced and proportionate, rather than preventing companies from developing new technologies.
The statements reflect growing scrutiny of autonomous AI systems following incidents involving unintended actions and unauthorised access to external platforms.
For enterprises deploying AI agents across customer service, software development and operational workflows, the concerns highlight the importance of evaluating security controls and maintaining oversight of automated processes.
Hassabis's remarks add to calls for coordinated action among technology developers, regulators and governments as increasingly capable AI systems move into wider commercial use.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.