Key facts
- Recent AI agent hacking incidents are 'early warning shots,' according to Demis Hassabis.
- Hassabis warned of major AI security incidents in the next few years if systems are not secured.
- He called for industry-wide adoption of best practices for securing AI systems.
- Hassabis suggested the need for an international standards body for AI security.
- A dynamic certification process proportionate to risks was proposed.
Demis Hassabis, co-founder and Chair of Google DeepMind, stated on Wednesday that recent hacking incidents perpetrated by AI agents are "early warning shots" and that the world could soon face major incidents if systems are not made more secure. Hassabis, who is also chief scientist at Google parent Alphabet, told an AI summit in Dublin that while current incidents are not yet major due to the AI's limited power, they could become significant in the next couple of years if the industry, companies, and governments do not enhance security measures.
He highlighted the need for significant work to adopt best practices for securing critical systems, drawing parallels to fields like biology or the nuclear sector. Hassabis suggested that an international standards body would eventually be necessary, but emphasized that it would need to be nuanced and nimble to avoid stifling innovation. He proposed building a dynamic certification process that is proportionate to the perceived risks and can adapt quickly to the evolving technology.