Inside AI Policy

May 15, 2024

Nonprofit developers tell NTIA of safety, cybersecurity risks from closed foundation models

By Mariam Baksh / April 10, 2024

Open-source researchers at EleutherAI have submitted comments to the National Telecommunications and Information Administration describing how generally closed foundation models can present the same -- or even greater -- safety and security risks than generally open models.

“Even for use-cases that are allegedly forbidden to a powerful closed model, it is easier to get a high-quality generation from a closed model via jailbreaking than it is to get a comparatively high-quality output from a weaker open model with no safety...


Log in to access this content.


Not a subscriber? Sign up for 30 days free access to exclusive news and analysis on artificial intelligence regulations and more.