Arc, an open SQL‑native time‑series database, contains an authenticated local‑file read flaw that allows a user with legitimate credentials to extract arbitrary files from the host. The vulnerability arises because the user‑SQL validator only blocks a narrow set of DuckDB I/O functions, while the broader set of functions such as read_csv, read_json, and glob are not restricted. RBAC checks that inspect FROM and JOIN clauses are bypassed because scalar table functions used in the SELECT list do not trigger table‑level enforcement. The flaw thereby permits reading sensitive files, violating confidentiality and potentially exposing credentials or configuration data, as identified by CWE‑200, CWE‑22, and CWE‑918.

The affected product is Basekick‑Labs Arc. Versions prior to 26.06.1 are impacted; all releases 26.06.1 and newer contain the structural sandbox that blocks file access outside the allowlist. Users of older releases should verify their installed version against this fix.

The CVSS score is 7.1, indicating high risk. Exploitation requires authentication and requires that the attacker is a valid Arc user; no remote unauthenticated vector is present. EPSS data is not available and the vulnerability is currently not listed in CISA’s KEV catalog, so the likelihood of exploitation is uncertain, but the potential impact on confidential data is significant. Once patched, the database will refuse to open files outside the allowed list and reject further INSTALL/LOAD operations, preventing the flaw from being reused beyond pre‑loaded extensions.