ArcadeDB DEFINE FUNCTION Scripting Authorization Bypass

Technical Analysis

The vulnerability (CVE-2026-67341, CVSS 9.8) stems from missing role/permission enforcement on the DEFINE FUNCTION SQL statement when LANGUAGE js is specified, allowing low-privileged or unauthenticated-but-connected users to register and invoke arbitrary JavaScript functions within the database engine's embedded JS runtime. Successful exploitation grants code execution in the context of the database process, enabling data exfiltration, lateral movement, and potential host compromise depending on JVM sandboxing and OS-level permissions. Because ArcadeDB is a multi-model database frequently used as a backend for graph/vector data stores and RAG pipelines, this flaw is especially dangerous where agent frameworks maintain persistent database connections with elevated service-account credentials. An attacker who compromises an AI agent's RAG data layer or gains any authenticated access to ArcadeDB could pivot to full RCE, exposing embedded API keys, vector store contents, and downstream LLM tool-call credentials, and could tamper with retrieved context to manipulate agent outputs (prompt/data poisoning).

Affected Systems

ArcadeDB Server and Studio, all versions prior to 26.7.2, including deployments used as graph/vector databases for RAG and AI agent backends

Indicators of Compromise

• N/A - authorization bypass logic flaw, no public exploit signatures released yet
• Monitor for anomalous DEFINE FUNCTION ... LANGUAGE js statements in ArcadeDB query logs
• Unexpected JavaScript function definitions by non-admin accounts

Remediation Steps

1. 1

   Upgrade ArcadeDB

   Update all ArcadeDB Server and Studio instances to version 26.7.2 or later where authorization checks are enforced on scripting functions.

2. 2

   Restrict Database Access

   Limit network and credential access to ArcadeDB instances to only trusted services and administrators until patched.

3. 3

   Audit Query Logs

   Review logs for DEFINE FUNCTION statements using LANGUAGE js executed by non-administrative accounts to identify potential prior exploitation.

4. 4

   Rotate Credentials

   Rotate database credentials, API keys, and service account tokens used by connected applications or AI agent pipelines in case of prior compromise.

5. 5

   Harden Agent Integrations

   For RAG or agent systems using ArcadeDB, enforce least-privilege service accounts and network segmentation between the database and downstream LLM tool-use components.