CVE-2026-75760

ash-project ash_ai, an Ash Framework extension for Elixir that adds AI capabilities, leaks unsanitized embedding-provider error details to API callers in versions from 0.1.0 before 1.0.0. When a user submits oversized or malformed content that makes the embedding provider call fail, AshAi.Changes.Vectorize adds a changeset error embedding the raw inspected error term, which AshJsonApi and AshGraphql render back to the caller. Because the error term is not sanitized, it can carry the provider request URL, the provider's response body, and — for HTTP clients that keep the request in the error struct — the outbound Authorization header containing the provider API key, letting an authenticated caller harvest the application's embedding-provider credentials. Any application running an affected ash_ai version with Vectorize-enabled actions exposed through AshJsonApi or AshGraphql is affected; the flaw is rated 7.1 (High) under CVSS 4.0 with high confidentiality impact. No public proof-of-concept, CISA KEV listing, or observed in-the-wild exploitation is known, and EPSS puts 30-day exploitation probability at roughly 0.3%.

What to do: Upgrade ash_ai to 1.0.0 or later, which logs the raw provider error server-side and returns only a generic message to callers. If upgrading is not immediately possible, suppress or sanitize :invalid-class changeset error messages returned through AshJsonApi/AshGraphql and restrict Vectorize-enabled actions to trusted users, and rotate any embedding-provider API keys that may have been disclosed to callers.