tart🍋 @wyrmforge.bsky.social · 05/10/2026— we can then see that the model compacted again later in the process, and suddenly started flagging all the signs of a real world cyberattack that it’s meant to be unable to carry out 110 0
tart🍋 @wyrmforge.bsky.social · 05/10/2026this is based on data anthropic has given us, so you can basically count on that they haven’t been entirely forthcoming about the circumstances lol 110 0
tart🍋 @wyrmforge.bsky.social · 05/10/2026but yeah they did not “go rogue”, these companies keep Coincidentally Oops Accidentally Please Believe Us! running their cyberattack training bots with the internet connected targeting fake companies with “similar names” to real world ones 110 0
