Only the latest commit on the main branch is actively maintained and receives security fixes.
| Version | Supported |
|---|---|
| main (latest) | Yes |
| older commits | No |
Public disclosure (non-sensitive issues): Open a GitHub Issue and apply the security label. Please include a clear description of the issue and steps to reproduce.
We ask that you avoid publishing or sharing vulnerability details publicly until we have had a reasonable opportunity to address the issue.
We operate on a best-effort basis. Our goal is to acknowledge reported vulnerabilities within 48 hours and provide a status update or resolution within 14 days, depending on complexity.
CSV DataLab accepts Anthropic API keys entered directly in the browser UI. The following describes how keys are handled:
- Keys are transmitted from the client to Next.js API routes via HTTP request headers on a per-request basis.
- Keys are never logged, written to disk, or persisted in any server-side store, database, or cache.
- Keys are never included in server responses or exposed to other users.
- The server uses the key only to forward the request to the Anthropic API and then discards it.
- Use scoped API keys with the minimum permissions required.
- Set usage limits and spending caps on your Anthropic account to limit exposure if a key is compromised.
- Treat your API key like a password. Do not share it or commit it to version control.
- Rotate your key immediately if you suspect it has been compromised.
The following are considered potential vulnerabilities:
- A user's API key being logged, stored, or exposed to any party other than Anthropic.
- Cross-site scripting (XSS) or injection attacks within the application.
- Server-side request forgery (SSRF) or unauthorized outbound requests.
- Any mechanism that allows one user's data or keys to be accessed by another user.
- Exposure of internal infrastructure details or credentials via error messages or responses.
The following are expected behavior and not considered vulnerabilities:
- The application requires a valid Anthropic API key to function — unauthenticated requests to the API routes are rejected.
- API keys entered in the browser are visible in the browser's memory and developer tools for the duration of the session. This is inherent to client-side key entry and is documented behavior.
- Usage costs incurred by intentional, repeated requests made by the key owner.
- Rate limiting or errors returned directly from the Anthropic API.