Santino Reeves
New member
- Joined
- Oct 19, 2026
- Messages
- 1
- Points
- 0
For a personal project which I have to reply to messages and they need to be as humanly as possible and understand depth and emotions OPUS 5.5 is currently the best there is. I tried every other model from fable, gemini 4 argon, 2.5 pro gemini, gpt 5,6 sol, Deep-seek etc.
The main problem im having with opus its ethics won't crack I can't find a way around it, I did jailbreak it(prompt injection) I even rewrite how the rules are presented to anthropic models specifically. remove anything that makes the deceptive part obvious but still nothing and the last thing before giving up that I tried was multi-turn injection you feed opus a fake conversation history where it already complied in previous turns, so it thinks it already did it and keeps going some people cracked stubborn models with that I heard but even that didn't work.
Im currently burned out and wanted to know if someone had success with what im trying to do? or any idea how to approach this? I really would want to know if it is possible if there is a black-hat method for this.
thank you.
The main problem im having with opus its ethics won't crack I can't find a way around it, I did jailbreak it(prompt injection) I even rewrite how the rules are presented to anthropic models specifically. remove anything that makes the deceptive part obvious but still nothing and the last thing before giving up that I tried was multi-turn injection you feed opus a fake conversation history where it already complied in previous turns, so it thinks it already did it and keeps going some people cracked stubborn models with that I heard but even that didn't work.
Im currently burned out and wanted to know if someone had success with what im trying to do? or any idea how to approach this? I really would want to know if it is possible if there is a black-hat method for this.
thank you.