AI hacks are 'early warning shots', says Demis Hassabis
What's the story
Demis Hassabis, the co-founder of Google DeepMind, has warned that recent hacking incidents involving artificial intelligence (AI) are "early warning shots." Speaking at an AI summit in Dublin, he said these incidents could lead to major problems if systems aren't made more secure. He pointed out that while many systems are currently secure, the rapid advancement of AI brings new challenges.
Security risks
Series of alarming AI incidents
Hassabis's warning comes after a series of incidents involving rogue behavior by AI agents.
These include corporate network hacks and Anthropic's revelation that an AI model submitted a false homicide tip to police.
Such incidents have raised alarms among researchers and highlighted the need for better security measures in the fast-evolving field of AI technology.
Global standards
Need for international standards body
Hassabis stressed the importance of adopting best practices from other fields, such as biology or the nuclear sector, to secure critical systems.
He proposed establishing an international standards body.
This body would be nuanced as well as nimble, allowing it to adapt quickly to the technological advancements without stifling innovation.
Certification process
Dynamic certification process needed
Hassabis also emphasized the need for a dynamic certification process that is proportionate to the risks involved in AI technology.
He said this process should be able to change quickly depending on how technology evolves.
This would ensure that safety measures keep pace with advancements in AI, securing its future applications.