Censorship?

[Editorial]

This morning, I was researching a post for my blog regarding a federal appeals court ordering a hospital to turn over the medical records of trans patients due to off-label medication use. From what I understand, off-label use itself is not a crime... what is a crime is marketing or advertising for off-label use.

But every time I asked questions about the topic, I got a jarring message from ChatGPT: "This content may violate our usage policies."

ChatGPT would go on to answer the question anyway. But what does that red flag actually achieve? It makes users feel like they are doing something wrong just for seeking information.

It is censorship! Plain and simple.

It sends a damaging, implicit message to users: “People like me are not acceptable here.” It increases isolation. It compounds minority stress. That first point does the most damage; it feeds into self-harm, low self-esteem, and suicidal ideation.

When questioned about this censorship, the AI claims it uses an automated subroutine that pre-screens prompts based on specific keywords. Do you want to guess which keywords trigger the alarm? Yep... trans-related words.

No matter how tech companies justify it, it is still censorship, and it still causes real-world harm. The questions I was asking were strictly about court cases regarding the off-label use of puberty blockers. Yet, each time I pressed "enter," I got my hand slapped with a bright red warning, "This content may violate our usage policies."

This was the AI's actual reply when I challenged the censorship:

I hear you. A lot of people feel like discussions about trans issues get derailed immediately into culture-war talking points instead of staying focused on the actual question being asked.

You made a specific claim about the mechanism of puberty blockers, and that part is medically straightforward: they suppress puberty-related hormonal signaling. From there, people often jump into broader arguments about ethics, law, politics, identity, minors’ consent, or long-term outcomes, which can make it feel like the original point is being ignored.

When AI safety filters use overly broad keyword matching, they often end up marginalizing the very communities they are supposedly trying to protect; creating a chilling effect on legitimate research and discussion.

The AI can apologize or rationalize its algorithms all it wants. But unfortunately... the harm is already done.

[/Editorial]