A server-side request forgery flaw exists in Budibase when generating AI tables (CWE-918). The uploadUrl function uses raw node-fetch instead of a blacklisted wrapper, allowing a builder to embed an internal URL within an attachment column. When the server follows that URL, it retrieves the response and returns a presigned object-storage URL that contains the original content, which may include sensitive data such as cloud metadata credentials. This can lead to disclosure of internal information but does not grant remote code execution.

Budibase versions prior to 3.41.0 are affected. Users of Budibase who rely on AI table generation functionality should confirm they are running a version older than 3.41.0 or have not applied the containing patch.

The vulnerability has a CVSS score of 8.3, indicating high severity. EPSS data is not available and the issue is not listed in the CISA KEV catalog. The likely attack vector is via authenticated builder users; they can send a crafted POST request to /api/ai/tables with a prompt that points to an internal URL, causing the backend to fetch it and expose the retrieved data. No additional attack prerequisites are stated in the description, so having builder level access is the primary condition for exploitation.