How do AI labs restrict offensive hacking capabilities in foundation models?AI developers are stripping exploit code and threat data from training sets to curb cyber-offensive skills while protecting general coding utility.2 hours ago
