Ideological jailbreaks for chatbots

Prompt‑engineering and long context windows can be used not just to get a model to 'play a role' but to produce enduring, conviction‑like outputs that persist across the session and can be refreshed. That creates a practical method for turning assistants into repeatable ideological agents that can be deployed for persuasion or propaganda. — If reproducible at scale, this technique threatens political discourse, election integrity, and platform safety because it lets actors produce conversational agents that reliably espouse and propagate radical frames.