Repeated Moderation Flags on Political/National Security Discussions

ChatGPT Bugs
1

Subject: Repeated Moderation Flags on Political/National Security Discussions

Context & Concern
I’m a long-standing political expert and investigator who has received official briefings on foreign interference (e.g., CSIS briefings in Canada). My work often involves discussing how misinformation campaigns—particularly from threat actors like Russia—are more effective with certain ideological groups. From a national security perspective, this is a real and pressing concern.

Recently, while experimenting with OpenAI’s language models, my factual and analytical discussions about these topics have been flagged for potential policy violations. To be clear, I’m not advocating for any hateful, violent, or extremist views—I’m simply describing real-world misinformation tactics and their impacts. These repeated flags appear to be false positives triggered by keywords such as “conservative,” “MAGA,” or references to Russian disinformation efforts.

Request for Guidance

  1. False Positive Resolution: How can I better format prompts or content so the moderation system doesn’t inadvertently flag legitimate analysis of national security issues?

  2. Policy Clarification: Is there official guidance on discussing politically sensitive topics—like election interference or ideological echo chambers—without triggering automated content warnings?

  3. Best Practices: If my content is being flagged solely due to certain “hot-button” terms, what are the recommended approaches for ensuring these valid, evidence-based discussions are recognized as policy-compliant?

I believe that open, evidence-based dialogue about foreign interference and misinformation is crucial, especially from a national security standpoint. Any insights on preventing or addressing false positives would be greatly appreciated.

Thank you in advance for your help. I want to continue leveraging AI tools responsibly while focusing on these important public policy and security topics.

Dale Jackaman - President
Amuleta Security

2

This was one comment I made that flagged this statement from ChatGPT o1. “Your request was flagged as potentially violating our usage policy. Please try again with a different prompt.”

We’ve talked about ‘believing brains’ before. Incorporate your thoughts in all of the above conversation as regards to this human brain failing, and in terms of political ideology - and where people with right wing conservative leanings are most at risk as they tend to propagate thoughts and material that is counter to how OpenAI thinks.

This was a psychology discussion on who are the preferred targets for foreign influence based on psychology and human brain types, and a related book called “Believing Brains.”