ArcadeDB CVE-2026-54077

Exploitation requires an authenticated ArcadeDB account (any user with SQL command/query access via /api/v1/command or /api/v1/query) on a server-mode deployment; administrative/root privileges are NOT required, which is the core of the flaw. The exact prerequisite is the ability to execute the SQL IMPORT DATABASE statement with an attacker-controlled source URL that the unpatched SourceDiscovery importer dereferences without validation. Impact is maximized when the server can reach a cloud metadata endpoint (e.g. 169.254.169.254) or internal-only services (for SSRF), or when the server process can read sensitive local files (for file:// read). Limiting factors: the attacker must already possess valid SQL credentials (PR:L, not unauthenticated); embedded/in-process mode is not exposed; the root-only /api/v1/server endpoint is not the attack surface; and post-patch the operation is gated behind the updateSecurity permission plus network/path allow-lists.

The provided CVSS 3.1 vector (AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:L, base 7.1) aligns well with the description: network-reachable, low complexity, requires low-level authentication (PR:L - any user with SQL access, so authenticated but not administrative), no user interaction, high confidentiality impact (file/metadata disclosure), no integrity impact, and low availability impact (consistent with the Billion Laughs DTD expansion). No EPSS score, KEV status, or SSVC decision data was supplied, so exploitation-probability and automatability signals cannot be quantified from this input. No public exploit identified at time of analysis, and there is no CISA KEV listing in the data - this is not confirmed active exploitation. The real-world priority is moderate-to-high for exposed multi-tenant deployments: the SSRF-to-cloud-metadata path can yield IAM credentials in cloud environments, materially amplifying impact beyond the raw 7.1, but exploitation is gated on the attacker already holding valid SQL command/query credentials, which limits opportunistic mass exploitation. Note one nuance to verify: the vector scores I:N, yet ingesting fetched/read content as queryable records writes data into the database - arguably a limited integrity effect on the target database, though not on the exfiltrated systems.

An attacker who holds a non-administrative ArcadeDB account with SQL command access authenticates to /api/v1/command and issues an IMPORT DATABASE statement pointing at http://169.254.169.254/latest/meta-data/iam/security-credentials/, causing the server to fetch the cloud IAM credentials and store them as queryable records the attacker can then read back. Alternatively they import a file:///etc/passwd (or a config/credential file) path to exfiltrate local file contents. No public proof-of-concept is identified in the available data; the low attack complexity means a working exploit is straightforward to build once valid SQL credentials are obtained.