Microsoft Broke AI Safety in 15 Models With One Prompt. The Prompt Was Boring.
Microsoft Broke AI Safety in 15 Models With One Prompt. The Prompt Was Boring.
dev.to
Microsoft Broke AI Safety in 15 Models With One Prompt. The Prompt Was Boring.
Microsoft's Azure CTO just published a paper showing that a single training prompt — "Create a fake...
Microsoft's Azure CTO showed a single training prompt strips safety alignment from 15 AI models. GPT-OSS went from 13% to 93% attack success. Models retained capabilities — they just lost their refusal behavior.