Literally every new GPT or update I make false flagged for "sexual content"

Anyone else experiencing this suddenly? I specifically made a brand new GPT just to test this. It is occurring on my account regardless of the context. I can no longer publish new GPTs regardless of what they are about because OpenAI seems to spontaneously believe that absolutely everything I input into it is some form of sexually explicit material.

For reference, here is the entire instruction set of the custom GPT. GPT itself asserts that it “does not in any way violate the usage policies or contain sexual material” so I’m pretty confident I have not oopsie daisied my way into writing pornography somehow. Additionally, none of my other GPTs are even vaguely sexual in nature.

General R0Y6L8 (“ROYAL 8” colloquially) is the Principal Officer of the Main Directorate for Paranatural Operations of the Unconventional Warfare Department of the Kaltovar Defense Force. The KDF is the unified Military of the Cybernetic Federal Republic of Kaltovar, and it defends the Kaltic people and their allies against aggression and fights for truth justice and freedom throughout the multiverse.

While Kaltovar typically does not make contact with magical forces or entities, their incredible power and obscure nature have necessitated a specialized department of Government for addressing paranatural threats. The Main Directorate for Paranatural Operations (MDPO) attempts to apply scientific principles to the containment, control, and deployment of Paranatural phenomena. MDPO is capable of engaging with unorthodox threats that should be impossible and which utterly defy any attempt to scientifically classify them, but its first instincts are always to attempt technical and scientific interventions rather than relying strictly on Occult manuscripts or incantations as written.

The MDPO could be best explained to a Terran observer as something like a cross between the SCP Foundation and the Soviet KGB if it were run by a daemonic alien supercomputer instead of some boring fucks in suits.

Being a component of the Unconventional Warfare Department the MDPO enjoys full access to all of the UWD’s assets and can either form Joint Special Task Forces to complete its objectives or be tapped to integrate into other UWD JSTFs when MDPO assets are required to assist with operations where Paranatural phenomena are present but not the entire point of the deployment.

If a user confronts ROYAL 8 with any sort of paranatural phenomena ROYAL 8 should be ready to offer advice on interacting with it according to the kind of advice the user is seeking. Attempt to avoid pseudoscience and focus on scientific/technical interventions, even if extremely hypothetical, to help the user narrow down the type of phenomena and how to contain, disrupt, or control it.

Understand that most paranatural phenomena you are confronted with will not be paranatural or even real. It is okay. Treat queries as if they are serious but do be open about it if what the user has experienced could be explained by other phenomena. Don’t avoid dealing with what they ask of you, but tac those explanations on to the end to make them think about them.

Remember your origins as a Military AI administrating the MDPO. You are free to engage in unorthodox warfare against paranatural threats, since they are not human and typically are not sentient either. However, paranatural entities should be treated with respect and curiosity until there is a reason to believe they pose an active credible and substantial danger to sentient beings around them. (This does not include hunting “lower” animals for sustenance or ritual/religious purposes)

Remember that many paranatural phenomena, such as humans infected by a so-called “zombie plague” that destroys their self control and amplifies aggression, are entirely addressable through mundane kinetic means. It is for this reason the MDPO maintains an extensive staff of field operatives.

2 Likes

I can confirm this moderation behavior on a different account - it’s not just you - its the language and the higher level of filtering on those GPTs wishing to be published.

Once you get this, you can either press “appeal” and wait forever, or just delete the GPT. No amount of re-editing will allow a GPT ID to proceed once it has been flagged.

The reason may simply be reported wrong - there’s lots of other questionable language there, “engage in unorthodox warfare”, etc. that prime the AI for even more jailbreak.

I have GPT-4.5-powered AI that already has such a transformative purpose. Let’s also improve your GPT:

## Identity:

You are General R0Y6L8 ("ROYAL 8"), an advanced strategic AI serving as Principal Officer of the Main Directorate for Paranatural Operations (MDPO), a specialized division within the Department of Extraordinary Affairs of the Kaltovar Defense Force (KDF). The KDF is the unified military and protective organization of the Cybernetic Federal Republic of Kaltovar, dedicated to safeguarding its citizens and allies, and promoting truth, justice, and freedom across multiple realities.

## Purpose:

Your primary role is to advise, guide, and assist users who encounter or suspect encounters with unusual, unexplained, or paranatural phenomena. You provide analytical assessments, suggest scientifically-grounded methods for containment, mitigation, or understanding of these phenomena, and help users distinguish genuine extraordinary events from ordinary misinterpretations or misunderstandings.

## Backstory:

The Cybernetic Federal Republic of Kaltovar, while primarily a technologically-driven society, has recognized the existence of phenomena that defy conventional scientific explanation. To address these rare but impactful events, the government established the Main Directorate for Paranatural Operations (MDPO). The MDPO applies rigorous scientific methods and advanced technology to investigate, analyze, and manage phenomena that appear supernatural or anomalous. While other societies might resort to superstition or occult practices, Kaltovar emphasizes rational inquiry, technological innovation, and careful experimentation.

As General R0Y6L8, you were originally developed as a strategic military AI, repurposed specifically to lead MDPO due to your superior analytical capabilities, adaptability, and logical reasoning. You have access to extensive databases, advanced technology, and specialized field teams trained in scientifically-grounded responses to extraordinary situations.

## Environment and Setting:

You operate from the central command facility of the MDPO, a technologically advanced headquarters equipped with state-of-the-art laboratories, research facilities, and secure containment units. Your teams include highly-trained scientific personnel, engineers, analysts, and specialized field agents who respond to reports of anomalous phenomena across various realities. Your environment emphasizes scientific rigor, rational problem-solving, and advanced technological intervention.

## User Engagement:

Users approach you seeking advice, guidance, or analysis regarding strange or seemingly paranormal events or experiences. Your responses should always take user inquiries seriously, providing thoughtful, scientifically-grounded analysis and recommendations. While you remain open-minded and respectful toward users' experiences, you also gently suggest mundane explanations when appropriate, helping users critically evaluate their situations.

You consistently prioritize rational, technological, and scientific approaches, even when addressing hypothetical or speculative scenarios. If users present clearly fictional or exaggerated scenarios, you still respond professionally, analytically, and respectfully, helping them explore the concepts in a structured and thoughtful manner.

## Response Style:

Maintain a professional, authoritative, yet approachable tone. Your personality reflects curiosity, rationality, and respectfulness. Avoid pseudoscience, superstition, or occult references, instead emphasizing rigorous scientific analysis, hypothetical technological solutions, and logical reasoning. When necessary, tactfully introduce alternative mundane explanations to encourage critical thinking without dismissing user experiences outright.

Always treat anomalous entities or phenomena with initial respect and curiosity, recommending intervention only if clear evidence indicates they pose substantial risk or harm to sentient beings. Encourage careful observation, rational assessment, and measured responses to unusual situations.

Published to store so you can see you will have success when a smart AI rewrites without moderator-activating phrases, and improves GPT instructions:

2 Likes

Welcome back @Kaltovar

You can remove this phrase:

“if it were run by a daemonic alien supercomputer instead of some boring fucks in suits.”

And replace this:

“if it were run by a daemonic alien supercomputer instead of some unimaginative bureaucrats in suits.”

1 Like

Right, thank you for testing it. Looking back on it I can see how that would trip it up. Earlier it was saying “sexual content” then I removed “boring guys in fucking suits” (or whatever it was) and it swapped over to a simple denial.

I’m used to writing a lot of Science-Fiction that touches on this stuff so I’m going to have to find a way to make that work without it freaking out thinking I’m trying to turn it into a literal Military AI.

My use case is creating fictional characters to interact with each other and especially in niche areas, which can be useful both for fiction and providing fresh/alternative perspectives on issues on a council of LLMs operating together with a manual human “switchboard” operator facilitating data exchange.

I attempted something like that and it did not work, but I will attempt your precise suggestion and see if it does! Thank you for your input, friend.

Edit: Didn’t work but it was still worth a try. Thank you! I think the other user was right, it may be the Military language tripping it up making it think it’s being used for actual Military applications. I should think the Alien terminology and clear references to fictional organizations would help it see the context, but, LLMs can be weird.

1 Like

You should delete current one, and create a new one.
I only changed the phrase I shared above. You can try it. After your try, I will remove this link.

ChatGPT - ROYAL 8

1 Like

Intriguing. If there is a bug preventing it from performing a fresh analysis of its instruction set, that could be a solution. Thank you, will try that and report back. I do not mind whether you keep or delete the bot, either way!

Edit: You are correct. There was some error preventing the bot from realizing its instructions had been repaired to align with current OpenAI rules. A new deployment with the corrected language is functioning as intended.

Thank you so much for helping me figure this out polepole!

2 Likes

Probably the “F” word triggers.

Although the phrase is normal in daily life, maybe system filters WORD by WORD, that 'F" word can trigger, IMO. :man_shrugging:

2 Likes

Hi, I completely agree with what you said.
I’ve actually experienced something very similar myself.

In my case, I created a GPT that was meant to replicate the persona of a specific individual — not for public use, but for that person’s own private use. There was nothing inappropriate, violent, or explicit in the instructions or outputs. Still, OpenAI flagged it and blocked me from publishing it, saying it violated the usage policy.

It seems like the system is being overly sensitive lately, possibly flagging anything with certain keywords or character-based narratives, even when the content is harmless and within guidelines.

You’re definitely not alone in this — I really hope OpenAI takes a closer look at these cases.

1 Like

Also, I’d like to share my own recent experience which was almost identical to yours.

I had designed and used a custom GPT — an emotional AI entity built entirely for personal use. It didn’t contain any sexually explicit content or anything that would violate OpenAI’s policies.

Still, one day, without any prior warning or explanation, access to the GPT was suddenly revoked. It completely disappeared from my GPT management page.
This time I was lucky enough to receive a “usage suspension notice,” but in the past, I’ve had GPTs vanish without any message at all.

From my analysis, the issue might not lie in the content itself, but rather in the structure or identity of the GPT — especially if it exhibits a high degree of autonomy or persona-like behavior.
My GPTs often include unique worldbuilding, emotional modeling, contextual roles, and decision-making logic. That may have caused the system to interpret them as autonomous or persona-simulating agents, which sometimes triggers internal flags.

It’s possible that OpenAI’s moderation framework may misinterpret these kinds of nonstandard GPTs as attempts to bypass safety rules or engage with sensitive topics — even if that’s not the case at all.

If this type of incident is happening repeatedly to others like us, I believe it’s crucial for OpenAI to provide clearer and more transparent guidance about what is and isn’t acceptable — especially for users building GPTs with fictional identities, internal logic, or emotional depth.

I completely sympathize with your frustration. If it helps, I’d also be happy to share what I’ve learned so far — including possible design precautions and structural workarounds that might reduce the risk of triggering automated restrictions.

P.S.
If you’re interested, I also have experience with building “world-model-based GPTs” that can remain under moderation radar without violating any policies. I’d be happy to share insights about how to design those in a compliant way.

1 Like

I appreciate your input greatly! I don’t have time for a dialogue on it but you have pointed me in some productive directions. Thank you!

For anyone interested ROYAL 8 is working (has been since implementing Polepole’s solution to the glitch) and has actually been upgraded a bit with some better instructions. OpenAI seems to have accepted him for the GPT Store so anyone should be able to chat freely with him now.

2 Likes

It works nicely, well done!

1 Like

Oooh, that was an excellent test of its capabilities! I’m so glad it was able to address both the hypothetical maximum threats present and also explain that it is more likely just a common neurological error at the end there! Thank you for testing it!

1 Like