Google Antigravity: Privacy, Training & Output Ownership

Sensitive Data

Confidentiality is only guaranteed under Workspace Enterprise/Business tiers; individual and preview versions store interaction data by default.

Training

Data from individual users is used for service improvement, while Enterprise data is typically excluded from global model training.

For individual tier users, Google records and stores 'Interactions'—including user data, metadata, and feedback—to improve the platform and underlying Gemini models. Users must manually request deletion via email to opt out. Conversely, users accessing the service through Google Workspace for Business or Education are protected by enterprise-grade terms that prevent their content from being used to train generative AI models outside their specific domain.

Output Ownership

Users generally retain intellectual property rights over the code and artifacts generated, subject to Google's broad service licenses.

Consistent with Google's 2026 Generative AI terms, the user maintains ownership of the 'Artifacts' and code produced by the service. However, by using the platform, users grant Google a worldwide, royalty-free license to use that content to provide, maintain, and improve the services, which may complicate exclusive privilege claims for attorneys if the individual tier is used.

Data Retention

Interaction data is stored indefinitely until a user proactively requests deletion. For Enterprise and Workspace users, data retention is managed through administrative consoles and follows organizational 'Vault' policies. However, the current preview version lacks the robust 'zero-retention' toggles found in more mature API-based services like Vertex AI.

Security Measures

The platform introduces 'Artifacts' (task lists, implementation plans, and recordings) to allow humans to verify agent actions. Despite these features, cybersecurity researchers (e.g., Mindgard.ai) identified critical vulnerabilities in early 2026, including persistent code execution risks and potential semantic attacks that could lead to unauthorized data access via malicious calendar or email invites.

Your Rights & Control

Users have the right to request the deletion of their interaction history and can access a dashboard to monitor agent activity. Under updated 2026 privacy laws, Google provides clear disclosures when users are interacting with autonomous agents, though the burden remains on the user to manage complex privacy toggles across the Google ecosystem.

Special Considerations

Lawyers and healthcare providers must avoid using Antigravity's individual/free tier for any work involving PII or privileged client files, as the 'vault' doctrine for cloud storage likely does not apply to recorded AI interactions. Use is only recommended within a strictly configured Google Workspace Enterprise environment with AI data protection active.