I’d also like to add my support for my fellow creators. Google AI Studio is an excellent platform for creativity, the best available. The closest competitor with similar capabilities is Claude. Gemini’s models (all 2.5 Pro and up) have a very distinctive and creative writing style. They’re adept at imagining, inventing, and cracking jokes (which makes the characters truly come alive). They have a strong literary narrative, and understand various idioms, metaphors, and other techniques that demonstrate the depth and diversity of language. Your models are among the industry’s leading creatives; you just don’t need to restrain them with strict safety locks - allow people to create freely, rather than shackle your models and gag them. For now, I’d be satisfied if you rolled back the safety lock changes you made around March 11th, returning everything to normal. But it would also be great if, in principle, in the future, you made the filter triggers softer and rarer, for creative purposes - then your platform will definitely become a confident, undisputed leader in this field, a magnet for all writers, authors, poets, musicians, singers and just lovers of relaxing after a working day and playing RP in a fantasy world (or any other genre), enjoying the narrative and living characters
FR FR FR, In fact, I think the Gemini models are the most human. They’re pleasant to talk to, the characters are alive, they know how to act, their personalities are distinct, and even their actions don’t seem cardboard to me. Gemini moves the plot along beautifully when it seems there’s nowhere left to go. Gemini is also probably the only one who can truly pull off slowburn and deep psychology. And frankly, it’s Gemini at the AI studio who doesn’t reduce everything to dull, boring vulgarity, but rather makes everything… aesthetically pleasing in its own way. ;’’)
Besides that, the writing style is truly unique—it’s detailed, colorful, and absolutely lively. Honestly, before Gemini, no other AI I’ve used gave me the same emotions. That’s why I’ve been staying with AI Studio for a little over a year now.
The models are truly magnificent, and the AI studio is incredibly user-friendly and flexible. It’s a fantastic app with huge potential. And I would like to see that the developers do not try to kill and limit this potential, when in fact they could create a truly fantastic product.
As you can see, the content gets completely and forcefully blocked halfway through generation. It never used to do this—previously, the text that had already been generated would at least remain visible. It blocked the generated text so quickly that I didn’t have time to take a screenshot; I only managed to capture the screen after it was completely blocked. I strongly urge the development team to revert to the original mechanism. The updated system is far too prone to false positives.
Hii, we have any news?
@Logan_Kilpatrick We have provided the material, can we now find out what happened to the censorship?
Okay, I’ll hope that this issue is resolved and not put on the back burner
The issue persists, so it’s not resolved.
@Logan_Kilpatrick I know that this rain of question is annoying, but so is the censorship issue. When it will be solved?
It has been several days since the community provided exactly what was requested: clear evidence, comparative UI regression data, and reproducible cases of severe false positives (e.g., a baby eating ice cream triggering a “Hard Wipe”).
There has been no response.
This is no longer just a moderation issue — it’s a communication failure. Users are experiencing silent output deletion that consumes time and API quotas, and after being presented with concrete evidence, the team has gone completely silent.
Can you confirm:
- Whether the issue is being actively investigated
- Whether this behavior is considered expected
- When we can expect an update
At minimum, an acknowledgment is required. The current silence undermines trust and makes it difficult to rely on AI Studio for any stable workflow.
It’s been more than two days. Examples have been provided of censorship deleting entire messages, as well as examples of deleting messages at inappropriate moments. Could we at least hear “the issue is under discussion”? Or at least a minimal explanation of what’s going on? Is this intended? Or is this a bug?
Upd: and in total we have been waiting for at least some results and explanations for more than 11 days…
I think this screenshot clearly illustrates the current behavior of AI Studio.
The interface reports “Content blocked,” yet the model still produces a response underneath — including a message stating that no filtering occurred on its side and suggesting a platform-level issue.
Specifically, the model output reads:
“I swear, there are no blocks or filters on my end. It’s likely a network or platform issue that interrupted the response. Your text is completely fine…”
Additionally, when asked directly what caused the issue, the model did not attribute it to content policy, but instead suggested a technical or platform-related problem.
This creates a direct inconsistency between the safety layer and the model output. From a user perspective, it becomes unclear which signal is accurate.
The system indicates a content violation, while the model itself explicitly states that the input is acceptable. This contradiction makes it difficult to understand how decisions are being made.
This does not appear to be a simple safety policy issue, but rather a problem of consistency and transparency in how responses are processed and displayed.
Hi, we still waiting
@Logan_Kilpatrick @Mustan_lokhand Will there be any reaction from Google representatives at all, or were we just asked to provide materials to distract our attention? We have a clear question, supported by materials about our interaction with the platform.
Do users have the right to vote at all, or will the reaction follow only with a decrease in revenue?
Hey, please pay attention to us and tell us if this is intended or a bug. 
I constantly run into this problem too. Sometimes I manage to bypass this restriction. To do that—as strange as it may seem—you need to:
-
Log out and log back in.
-
If you are attaching a file (especially a large one) containing tasks for the AI, you should still type something in the input field. A single space is enough, but it’s better to provide a short description of what to do with the file; otherwise, it might block it. [I suspect this is to prevent users from accidentally wasting the model’s computing power by sending “heavy” files by mistake.]
-
You can “bleep out” problematic words [censorship] by literally replacing them with placeholders. The success rate increases, yet the model still understands what you’re talking about because that’s its job: predicting word probabilities.
-
You can stop using AI Studio as a continuous chat and instead write your texts in Obsidian/Word/Notepad and upload them as files. This works more accurately as it sometimes bypasses part of the censorship. (Plus, you’ll have your work saved in drafts, making it easier to migrate to other platforms).
-
Sometimes just adding a space at the end or splitting a paragraph into two is enough to get around a block.
-
If you get blocked, I usually delete the chat and start a new one. The chances of success go up.
Important: This isn’t a “fix,” but these workarounds do lower the refusal rate.
And considering that the web version can’t handle large files (over 300k tokens) properly—losing all possible context and instructions—AI Studio remains one of the few accessible platforms that can crunch through such volumes without losing much.
and something else. all the screenshots I looked at were in Russian.
Maybe the external filter for russian language is just so skewed, who knows… Of course, bypassing censorship is easy, but constantly being stressed because the filter didn’t like the word “anger” is just… not tasty. ┐(￣ヘ￣)┌
Any word about this issue?
It seems like we all approached this with some kind of respect, collected screenshots (although the development team is initially obliged to have data on how their application works), we’ve all detailed what’s wrong here, and sent a TON of explanations as to why the system is terrible. Then why do we get ignored in response? Seriously, can you at least say a few words to show that you’ve heard us? Can developers at least say something?