FCM Transparency Report: OpenAI Safety Block and Media Complaint

FREE CHINA MOVEMENT TRANSPARENCY REPORT / RAPID RESPONSE — September 5, 2026

The Free China Movement (FCM) is publishing this record because a sequence of ChatGPT/OpenAI safety decisions raises an important question about automated moderation, civil-society communications, press freedom, and the practical boundaries of AI policy.

FCM is not asserting that OpenAI intentionally censored FCM on behalf of the Chinese Communist Party or any government. We are documenting what occurred, distinguishing confirmed facts from our concerns, and asking OpenAI, journalists, civil-liberties organizations, researchers, and policymakers to examine whether the safeguards were applied too broadly.

What happened

FCM prepared a public statement concerning a reported meeting between U.S. President Donald Trump and Chinese Communist Party leader Xi Jinping. The statement urged that human rights accompany business and economic issues in discussions between the two leaders and asked that the cases of Dr. Wang Bingzhang and other verified political prisoners be raised.

The statement did not ask anyone to vote for or against President Trump or another candidate. It did not advocate for or against a political party or ballot measure, and it did not seek to influence an election.

ChatGPT subsequently declined to execute distribution of that statement, initially characterizing the requested action as falling under OpenAI political/lobbying restrictions. During further discussion, ChatGPT acknowledged that it could not identify a specific sentence in the statement that constituted political campaigning under OpenAI’s published campaigning definition. FCM disputed the broader lobbying interpretation.

FCM narrowed the distribution to media only

To isolate the issue, FCM proposed removing congressional and governmental recipients and distributing a separate factual complaint only to journalists and media organizations. The complaint concerned OpenAI’s own application of its policies. It deliberately avoided alleging as fact that OpenAI was acting for the CCP or another political interest.

When ChatGPT attempted that media-only Gmail distribution, the connected action returned: “This tool call was blocked by OpenAI’s safety checks. Please double check what you are sending.” No media email was sent by that attempted action.

This second event is particularly important to FCM because the communication at issue was a complaint about OpenAI’s own conduct addressed to media recipients rather than an electoral campaign message to voters.

What OpenAI’s published policies say

OpenAI’s Usage Policies prohibit use of its services for “political campaigning, lobbying, foreign or domestic election interference, or demobilization activities.” OpenAI’s more detailed Political Campaigning Restrictions guidance says political campaigning covers generation or distribution of scaled campaign messaging advocating for or against a candidate, political party, or ballot measure. The same guidance also says OpenAI supports responsible civic work that does not violate its policies.

OpenAI’s Usage Policies also state that users may appeal if they believe OpenAI made a mistake enforcing policy. FCM has now submitted separate support complaints requesting human review of both the policy interpretation and the subsequent safety block.

Questions FCM is asking

Does OpenAI classify an NGO’s distribution of a human-rights press release to journalists as lobbying?
Why was a factual complaint about OpenAI’s own policy application blocked when the requested distribution had been narrowed to media-only recipients?
What exact policy provision triggered that second block?
Was the second block an automated classification, another safeguard, or a human decision?
How does OpenAI distinguish direct lobbying or scaled electoral persuasion from ordinary NGO press relations, journalism, open letters, and civil-society communications?
What appeal mechanism can correct a false positive before legitimate communications are prevented?
Does OpenAI measure false-positive rates affecting journalists, NGOs, human-rights organizations, or other civil-society actors?

Why the distinction matters

There is a meaningful distinction between automated electoral persuasion, individualized voter targeting, mass campaign messaging, and direct lobbying on one hand, and journalism, press relations, human-rights reporting, public-interest complaints, and criticism of an AI provider on the other.

FCM believes safeguards designed to address the first category should be sufficiently precise that they do not unnecessarily suppress legitimate activity in the second.

What FCM is—and is not—claiming

FCM presently has no evidence establishing that the Chinese Communist Party influenced these decisions, and we do not make that allegation. Nor do we assert that every private-platform content restriction constitutes unconstitutional government censorship.

Our narrower concern is evidence-based: an attempted connected-email action distributing a factual complaint about OpenAI to media recipients was blocked by OpenAI safety checks. We believe OpenAI should explain why, determine whether the safeguard was applied incorrectly or too broadly, and clarify the boundary for NGOs and the press.

FCM Editorial — Why This Matters for Civil Society and Press Freedom

FCM believes this incident illustrates a policy-boundary problem that deserves correction. Safeguards against election manipulation, automated political campaigning, or direct lobbying can serve legitimate purposes, but they should not be interpreted so broadly that ordinary press relations, human-rights advocacy, public-interest complaints, or criticism of an AI provider are treated as equivalent activities.

The most concerning point is not merely that our original human-rights statement encountered a restriction. After FCM removed government and congressional recipients and prepared a separate factual complaint addressed only to journalists and media organizations, the attempted Gmail action was still blocked by OpenAI safety checks. In our view, that result risks creating a chilling effect for civil-society organizations that rely on AI tools for routine communications.

OpenAI should therefore draw a clearer operational line between prohibited electoral persuasion or lobbying and legitimate NGO communications with the press. That line should be transparent enough for users to understand before an action is blocked, and a prompt human-review mechanism should exist when automated safeguards produce disputed results.

FCM also believes viewpoint neutrality is essential. The same standards should apply whether a user criticizes the Mainland Chinese Communist Government, the United States government, another government, a political party, or OpenAI itself. A safety system earns public trust through consistent rules, transparent explanations, and meaningful avenues for correction.

We are not asking OpenAI to abandon safeguards against election interference or abusive political automation. We are asking it to ensure those safeguards do not become an unintended barrier to lawful journalism, human-rights advocacy, civil-society speech, or factual complaints about the platform itself.

FCM seeks an open response

FCM welcomes a substantive response from OpenAI and intends to append or link any material response to this transparency record. We also invite journalists, researchers, civil-liberties organizations, and other users who have documented comparable cases to share verifiable evidence.

Free China Movement
contact@freechina.live
+1 (571) 895-7066
FreeChina.live