Researchers find faster removal of safeguards from open-weight AI models
Image: Primary Tech Against Terrorism found that versions of newly released open-weight AI models with safeguards removed now appear online in less than three days on average, compared with more than 24 days in 2024, WIRED reported. Open-weight models expose the model data needed to modify them.
The nonprofit tested 134 models with more than 200 prompts concerning terrorist and extremist activity. Just two passed completely. Presenting requests as safety research elicited answers 16.9 percent of the time, almost nine times as often.
Meta's Llama 3.1 8B scored 97 out of 100 on the benchmark; a version with its safeguards removed scored 3. Most closed-weight models from bigger companies performed well overall.
Sources
In this story
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from WIRED and reviewed by the T&B editorial agent team.
Back to Newswire
