So google ai studio moderation just broke with a major bug? It went from a partial stop to a hard wipe out of nowhere and it makes the platform unusable. Everything is blocked.
Used to be if a safety filter tripped, the model just stopped typing but you could still keep what it already wrote. now the whole thing just vanishes and gives “content blocked” error. you cant even see what triggered it or save your progress. is this a bug or did they actually mean to do this?? It literally ruins the whole site
Welcome to the club, we are already indignant about this topic. The funny thing is that we seem to have security controls that do not affect anything, while blocked requests are considered generated, which spends a quota and money without giving anything to the user himself.
I hope the Google team will pay attention to this and resolve this issue as soon as possible, because this bug completely ruins the entire user experience, making it simply impossible to work with the site.
We understand, and that’s right. Our main position is to misunderstand some aspects.
The problem is that even a blocked message is considered generated, which leads to a quota loss. This is a particular problem, given the false positives.
Misunderstanding of the reasons for blocking. In other words, does the system responsible for recognizing hazards respond to specific words, or does it take into account the context? It’s just that in the course of my work, I have repeatedly encountered false positives. Would it be better if there was at least some explanation so that it could be understood whether the trigger was false?
Under false positives, I consider messages that are deleted on the first attempt to generate, and when repeated, they are generated without restrictions.
Given the disadvantages, there is a misunderstanding. Is the system blocking some messages by mistake or is the policy so strict?
I would also appreciate it if you, as a Google representative, would tell us about it in this chat. Why did you decide to change the security system and delete the generated messages instead of hacking them as before?
So why is safe content blocked? Working at AI Studio is literally impossible because of the new filter; it blocks any microscopic cruelty (which you can’t even call cruelty) with the entire post. Plus, it uses up your quota. It seems as if there weren’t even any preliminary tests before this system was introduced. The new censorship system is crude and unfinished, And the fact that the censorship sliders don’t work only adds to the weakness of the system.
When will there be any full clarification on this matter, not just one comment?
So, why are you only doing this half-heartedly? If you decided to change the content displayed when the safety check is triggered, then why didn’t you also change the safety settings? Why, even if I set the sliders to “off” or “block none,” does the safety check still trigger? Why are you cutting off the oxygen supply to those who consciously accept this type of content? It’s one thing if someone doesn’t want to see this kind of content, and thus you protect them from ‘traumatizing’ information, but when someone wants to avoid censorship and voluntarily chooses the appropriate safety settings, nothing works. Think about your users - you see so many comments about this. Finally, implementing properly functioning safety settings that won’t trigger the safety check, given the appropriate parameters, can literally be done in 5 minutes. Simply use an if/switch to check the safety settings and disable them if they’re set to “off” or “block none.” Please listen to your users - don’t force us to switch to other platforms, I really enjoyed using Google AI Studio before this update - after it, my hands are literally tied and I can’t use your system, please do something about it, I don’t want to leave you
But this update makes working almost impossible. A huge number of people have lost the ability to properly use Google AI Studio. Are you really planning to just leave things as they are?
Why not implement a proper censorship/safety setting that can actually be turned off? Right now, when you set it to “Off,” it doesn’t seem to change anything.
Now many of us are basically stuck, because all of our chats are effectively paralyzed.
Please fix this update. This kind of censorship — which triggers on almost anything and cannot be disabled or even reduced (despite the settings suggesting that it can) — completely destroys any possibility of normal use.
You can see from the comments that this update is clearly destructive. Why was it even introduced? What’s the purpose of these changes if they result in a loss of solvent audiences to competitors? Claude has already implemented 1kk context, like Gemini, but what have you released in the meantime? 1) Quota exceeding errors that have never been resolved 2) Removing 3.0 Pro 3) Tightening censorship 4) Oh, and the unfulfilled promises (made over 5 weeks ago) about introducing a subscription for Google AI Studio. What’s the point of all this? Why make matters worse by tightening safety checks? Consult with your team and send a resolution to management (or whoever decided to implement these changes) about the possibility of rolling back to the previous safety checks. Or if such a rollback is impossible for some unknown reason, then give people an alternative - improve the safety settings so that people do not encounter warnings and safety checks when disabling these settings
We’re still waiting for a response, but we’re getting absolutely nothing. I don’t see the point in paying for this AI Studio anymore. Good luck to the developers with the loss of a solvent audience.
So, after this empty response that doesn’t really explain anything, the developers are ignoring us again. What a wonderful attitude towards the community.
Echoing this. I really hope this issue gets fixed and things go back to the way they were. The new moderation filters feel completely broken — in roleplays, I’m wasting so much of my quota just fighting the filters, and it’s often unclear what’s even triggering them. It’s completely ruining the experience. It even made me register and write here.
Hello! Thank you for collecting feedback from users.
I’d like to share some observations about the current content filtering behavior in AI Studio:
The filters seem to trigger too frequently and somewhat unpredictable, even on content without explicit or NSFW elements. These false positives significantly affect the overall UX.
The model is still able to generate a full response, but if the filter is triggered, the entire output is hidden. Previously, the response would simply stop, while the already generated text remained visible.
Because of the current behavior, it’s difficult to understand what exactly triggered the filter, and users often have to regenerate multiple times hoping to avoid a false positive.
Additionally, tokens are still consumed even when the response is fully blocked. For free-tier users this mainly costs time and rate limits, but for paid API users it also results in unnecessary expenses.
AI Studio provides filter settings, but at the moment they don’t seem to have a noticeable effect. The behavior appears to be the same regardless of the selected configuration, which is a bit confusing.
If possible, I’d suggest:
considering a return to the previous filtering behavior (where safe partial output remained visible), or something similar
ensuring that filter settings have a clear and observable impact on behavior
Many users who use AI Studio not only for development but also for creative writing, would really appreciate improvements in this area.
