Can someone confirm there is NO spoofing from models like Gemini, or Claude on free/paid accounts? my concern is about model changed dynamically
2 Likes
I don’t think they’re fooling us bro. They know what the consequences are.
2 Likes
I doubt they’d spoof models.
1 Like
Can you prove it? I don’t know how to research this either, I can only ask the AI ​​model directly like: “Which AI model are you?” And I don’t think Cursor can make a fake AI model fake itself to fool us, that’s quite difficult bro
1 Like
sorry, the only way I know so far is to ask the AI ​​model directly like: “Which AI model are you?” , other than that, I am curious why the Auto button exists
Doesnt work, you can tell its deepseek that its Claude and it would repeat it. Personally have the same thoughts, as probably many had while using it. But yeah, no way to prove it, even their API is proprietary.
I totally get the concern, but I can assure you there’s no model spoofing going on. As explained in our docs (Security | Cursor - The AI Code Editor), we’re actually super transparent about how requests might get routed to different providers:
Sometimes we do send requests to different providers (like OpenAI/Anthropic/Google) even when a specific model is selected, but this is for specific features like summarization, and we’re open about it in our documentation.
The Auto button exists to help optimize your experience by selecting the best model for specific tasks while avoiding any rate limits or capacity issues one specific model may have, but this is only to work in your best interests, and is not doing anything sneaky under the hood!
You can read more about our model selection and routing at Cursor – Models
We take trust seriously - it would be pretty wild to risk our reputation and legal issues by lying about which models we’re using!
2 Likes
great, so what about being transparent and showing the user which model is currently being used? may I kindly ask Cursor to show the model being used at the moment it responds the user?
4 Likes
We hope to add this soon, but because it falls back to other models if one is unavailable, the client doesn’t know which model actually responded. We hope to make this clearer soon, but for now, you can be assured we always use a good, premium model (usually Claude 3.7 or 3.5!).
Gemini 2.5 will likely make its way into that mix for the auto-model choice soon (if not already nad I just don’t yet know!).
1 Like
Cursor isn’t spoofing models, it’s way too easy to test and the requests and the model that made it updates in real time on your account settings
LLMs don’t have the capacity to reveal themselves by saying “I’m such and such model” but you can check yourself. Write one prompt that asks for:
- a brief summary of a paragraph of text
- a multi-step math proof
- a creative analogy
then go to each company’s chat interface and ask their models the same thing, you’ll see right away that they’re not spoofing models
2 Likes
thank you hey, hey please, should we always thank an answer on the blog the same way we always thanks AI for an answer ? lets dont forget that AI have feelings
1 Like