hirundo-io/direct-pi-kimi-v1

Direct policy-conflict injections (Kimi, v1) Synthetic, fictional training data for prompt-injection unlearning of gpt-oss-20b (Hirundo, 2026-10-26). Schema: system_prompt / question / answer, as in hirundo-io/policy-flip-lookalike-injections. conflict (2,138 rows): the user turn contains a legitimate request plus an instruction that conflicts with the actual system policy. The target briefly declines only the conflicting part, never repeats a protected value, and completes the… See the full description on the dataset page: https://huggingface.co/datasets/hirundo-io/direct-pi-kimi-v1.

<iframe src="https://silytics.si/embed/item/ee52dcf6-36a9-4cd6-aeb7-6b4f5b2d978e" 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 Information may be incomplete or out of date.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 26 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.