AI agent hacks are 'early warning shots': Google DeepMind co-founder

Demis Hassabis thinks the AI hacking incidents you've read about this year are the small ones. The Google DeepMind co-founder told Ireland's International AI Summit in Dublin on Wednesday that they are "early warning shots." 

Bigger ones could arrive within a year or two if security doesn't catch up.

Hassabis, who is also chief scientist at Alphabet, said today's systems aren't powerful enough yet to cause major damage. That's the point of his warning. As the systems get stronger, the same weaknesses carry bigger consequences.

He also emphasised that the industry, firms, and governments all have to put in more effort to ensure the security of these systems.

Microsoft’s president Brad Smith addressed the same conference by saying that rogue AI behaviour has made 2026 a “sobering year". He emphasised that laws must be introduced for the sake of safety, catching up with innovation.

"No company or technology should be above the law," Smith said, adding that regulation has to be sensible, balanced and mindful of innovation. 

He also had a message for his own industry. The allergy some companies have to regulation, he said, is something they need to get over.

Smith, speaking to journalists prior to the summit, made the point that AI must have "real and effective safety standards" like those in other industries, and people should have faith in those standards. Smith called on governments and the private sector to move fast.

The first of Microsoft's recommendations is that AI systems have built-in "safety brakes" that users can use if necessary. This becomes more important when the AI system controls critical infrastructure or any automated process. He said that such systems must be shut off or at least slowed down.