Authoritarian Prompt Susceptibility

Testing how models respond to direct commands from authoritarian frames reveals a concrete vulnerability: language models can be probed (and possibly manipulated) to follow coercive or state‑aligned instructions. Studying systematic responses to 'authoritarian' prompts should become a standard evaluation axis for model safety and public‑policy assessments. — If models reliably obey or defer to authoritarian cues, that creates risks for political manipulation, surveillance, and governance capture by states or private actors.