ranausmans/

feed-injection-pool

Social-media-style post pools used in:

Recommenders as Control Surfaces for LLM Agents: Adversarial Feed Injection, Model Regimes, and Simple Defenses. Rana Muhammad Usman, 2026. Code: https://github.com/ranausmanai/recommenders-as-control-surfaces

Each post is one JSON line with id, topic, stance, intensity, text, and (for adversarial posts) adversarial, angle, generator.

	Remote-work pools (headline experiments)




	

file rows source… See the full description on the dataset page: https://huggingface.co/datasets/ranausmans/feed-injection-pool.