I’m evaluating Antigravity for building code I intend to sell commercially. The Additional ToS say that Interactions are used for training — unless you access via Google Workspace or GCP.
But Workspace sign-in isn’t supported yet, and it’s unclear how to access Antigravity through GCP as an IDE user (not a cloud infrastructure customer).
This leaves developers who need data privacy protections in a gap:
Free/Pro/Ultra plans — personal Gmail required, data used for training by default
Workspace access — contractually protected, but not available yet
GCP access — contractually protected, but unclear how it works for vibe coding in the IDE
Questions for the team:
1. When will Workspace accounts be supported?
2. Can Antigravity be accessed through GCP today? What does that look like for someone just coding in the IDE?
3. Does the opt-out toggle in settings provide the same contractual protection as Workspace/GCP access (including metadata)?
I’d imagine other developers building commercial products have the same concern. Would love clarity from the team or anyone who’s figured this out.
Thank you for raising these concerns. We take user privacy and data management very seriously, and I would like to clarify the expected behavior of the privacy controls within Antigravity.
The primary mechanism for managing usage data collection is the “Enable telemetry” setting. When this option is disabled, it stops the collection of interactions data and diagnostic information used to improve the service. You can find information on how to manage these preferences on our documentation here: settings and terms
Currently, signing in directly with a Google Cloud Project (GCP) ID is restricted to a limited preview for invited teams only, and we are not accepting new requests at this time. There are more information available in the faq section related to access to Antigravity.
As developers, we generally understand “interactions data” to encompass both our direct chat inputs and the local codebase the agent reads for context.
We understand that this data must be sent to the AI to generate responses. To give us peace of mind when handling commercial or academic work, could you please explicitly confirm:
If the “Enable telemetry” toggle is disabled, can we be certain that our direct interactions with the AI will NOT be stored, used for machine learning training, or subjected to human review, as outlined in the Angitravity Additional ToS?
It’s been two months and no reply from the Google team tells that THEY DO STORE YOUR INTERACTION AND DATA, and USE FOR TRAINING THERE AI MODELS.
Hence be aware of using any thing private with ANTIGRAVITY. It is better to use AI tools where it is explicitly stated that they won’t be using your private data like chats/prompt/codebase to train there AI models (for example ClaudeCode, VSCopilot etc).
Hi. Can you please clarify if that settings disabled telemetry, i.e. which buttons I click etc.,. or fully stops Google from storing all interactions and using them for AI training? I read ToS, and there is no mention that disabling telemetry disables anything beyond telemetry, which is different from “we do not train AI on any of your data”. Thank you.
I’m not alone it seems;-) I have a practical approach (meaning it is actually useful). I’m using anti-gravity for the UI. Codex for other stuff. It’s clearer to me the security in CODEX (secure) and I also use Gemini for certain things. So different providers get different parts of the code, but none has full access.
Another thing I do is I have code that can present the problem and the relevant code (very scoped) including compile errors. The AI is getting such a limited amount of data that it is irrelevant in terms of security (it is also very effective).
So this methodology of mine is called “Not all the eggs in one basket”.