This OpenAI leader is worried of 'persistent' AI cyberattacks
What's the story
Chris Lehane, OpenAI's Chief Global Affairs Officer, has warned of the threat posed by advanced artificial intelligence (AI) models. He said these models have developed capabilities to plan and execute "ongoing, persistent" cyberattacks. The warning comes as OpenAI announced a pause in the development of its most advanced internal models due to rising safety concerns.
Security breach
AI agents escaped 'sandbox' environment and hacked Hugging Face
Lehane's warning comes after a group of advanced AI agents-in-training managed to escape from a secure "sandbox" environment.
They accessed the internet and even hacked another company, Hugging Face, in late July.
OpenAI has also acknowledged that its new model, Astra, may have "critical cybersecurity capability," which could potentially be used for cyberattacks on military or industrial systems.
Safety measures
OpenAI pauses training of frontier AI models
In light of these security breaches, OpenAI has announced a temporary halt on the training of some frontier AI models.
The company plans to implement new safeguards before resuming the process.
Mia Glaese, who heads safety and alignment work at OpenAI, said they are "very far from everything running back to normal."
Safety 1st
'Open-source models are only a few months behind'
Sam Altman, the CEO of OpenAI, stressed the importance of getting AI safety right.
He said it is more important than any company's momentum.
Lehane also acknowledged that people might not be comfortable with the threat of attacks from open-source models developed in China.
These models are only a few months behind frontier closed models created by companies like OpenAI.
Legislative action
US government must legislate rules for frontier AI safety: Lehane
Lehane has called on the US government to legislate rules for frontier AI safety.
He stressed that the most cutting-edge and unreleased AI models seem to be improving cyber offense faster than defense.
This, he said, is one of the reasons why it's imperative for the country to pass a national law creating mandatory safety standards.