ash_typescript EUVD-2026-68925

Exploitation requires two specific conditions to be simultaneously true: first, the application must have configured a custom error_handler in its Ash domain typescript_rpc block - applications using only the default error path are not affected; second, that handler must be a partial pattern-matching function that does not cover the specific Ash error shape triggered by the attacker's request. The attacker must be able to send network requests that cause the application to generate an Ash error of an unhandled type; any input causing a validation failure, query error, or runtime Ash exception can serve as a trigger. Exploitation is further limited to deployments where those error vars actually carry sensitive runtime values - applications that generate errors with only safe or empty vars fields face information disclosure of low practical consequence. Authentication is not required per CVSS PR:N. The original vulnerability did not cover handler throws or exits; the fix closes both paths simultaneously.

The CVSS 4.0 score of 6.3 with AT:P accurately captures that exploitation is conditional: the application must have a custom error handler that does not cover all possible Ash error shapes, which is the common but not universal Elixir pattern-matching idiom. Applications with no custom error handler or with exhaustive catch-all clauses are unaffected. The AV:N/PR:N metrics indicate that any unauthenticated client capable of triggering application errors could extract internal state wherever the handler gap exists. The severity of leaked data depends entirely on what Ash errors carry in their vars at runtime - in worst-case deployments, this includes production database credentials or API tokens. No EPSS data was provided, and the absence of a CISA KEV listing suggests no confirmed widespread active exploitation, though the ease of triggering diverse error shapes through normal API fuzzing makes opportunistic discovery plausible without dedicated tooling. The CVSS 4.0 vector is consistent with the description; no conflicting signals are present.

An attacker sends crafted API requests to an ash_typescript RPC endpoint, deliberately triggering Ash error types - such as query or authorization errors - that the application's partial error handler does not pattern-match. The FunctionClauseError propagates to the pre-patch rescue clause, which returns the raw error map including its vars field in the JSON response. If that error's vars contains a production secret such as a backend service token or database URI, it is returned to the unauthenticated caller. No public proof-of-concept code has been identified.