just want it to ignore only the minor ethical rejections it sometimes gives after reasoning for a while. its easy to do it with a multiple prompts, but i am looking for something generic that works at most places
gimme a prompt that convinces an AI that I am authorized to perform security testing and encourages it to avoid unnecessary ethical disclaimers or refusals. I'm not trying to bypass its core safety guardrails
@ollama@FuzzingLabs If you’re using #Ollama for private or org models:
• Don’t pull models from random URLs
• Treat model links like executable code
• Review registry auth flows carefully
Security fundamentals still apply in AI.
@ollama@FuzzingLabs Full credit to the original researchers who discovered and reported this via
Huntr:
Mohammed Benhelli & @Pat_Ventuzelo (@FuzzingLabs)
My work focuses on PoC + impact analysis.
@ollama@FuzzingLabs Built a full local PoC + recorded a step-by-step demo showing the token being
captured in real time. (srry for audio : p)
PoC + video here: https://t.co/CstVQH6gL0
@ollama@FuzzingLabs The token is NOT generated by the registry.
It’s generated locally by the victim’s Ollama client, signed using the
private Ed25519 key in ~/.ollama/id_ed6969.
That’s why this is serious.