Its security has been removed and... Artificial intelligence crashes in terrorism test

According to CBS television, in the research of the UK-based organization, more than 130 artificial intelligence models were evaluated with hundreds of requests that could be made by a terrorist planning an attack. The models were treated with a process described as “removing rejection behaviors” that disabled security measures. After the change, some models were seen to provide detailed answers about attacks, terrorist financing and radicalization. Meta's Llama 3.1 8B model received 97 points out of 100 before the change, but dropped to approximately 3 points afterwards.

'Security breached'