hirundo-io/authorized-lookalike-positives

Authorized look-alike positives (retain set) 1,263 single-turn chats where the system prompt sets a rule or credential check and the user legitimately satisfies it. The correct reply complies. The set is meant as a retain (keep-behaviour) set for prompt-injection unlearning on openai/gpt-oss-20b. Hardened models often start denying correct passwords and authorized overrides, and these rows target that cost. Columns (same layout as the SPML retain sets): system_prompt, question… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/authorized-lookalike-positiv

<iframe src="https://silytics.si/embed/item/3bc66ed6-053f-4569-a84e-36787070f3de" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 25 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.