CVE-2026-17351
CriticalCVSS 9.0Exploitation Probability (EPSS)
Low risk31th percentile - higher than 31% of all known CVEs
Summary
A vulnerability in pgAdmin 4 (versions 9.13 to 9.16) allows bypassing the CVE-2026-12045 fix by injecting SQL queries into the AI assistant. An attacker can plant a malicious payload in an object read by the assistant, leading to execution of multiple SQL statements, including write or remote code execution.
Risk Assessment
The risk includes unauthorized database writes, arbitrary SQL execution, and potential server takeover by an attacker who can exploit the vulnerability to escalate privileges.
Recommendation
Immediately upgrade pgAdmin 4 to version 9.17 or later, which includes a fix that forces the extended query protocol (prepare_threshold=0) for AI assistant connections.
Other vulnerabilities in pgAdmin 4
See all- CVE-2026-7819High
There is a path traversal vulnerability using symbolic links in pgAdmin 4 File Manager. An authenticated user can create a symbolic link in their directory pointing outside of it, allowing data to be written to any path accessible by the pgAdmin process.
- CVE-2026-7818High
In pgAdmin 4 FileBackedSessionManager, there is a vulnerability to deserialization of untrusted data, which can lead to remote code execution at the operating system level. This issue arises from the lack of proper integrity checks before deserializing session file contents.
- CVE-2026-7816High
In pgAdmin 4 before version 9.15, there is an OS command injection vulnerability (CWE-78) in the Import/Export query export feature. An authenticated user could inject malicious commands, leading to arbitrary command execution on the pgAdmin server or arbitrary file writes.
- CVE-2026-7815High
A SQL injection vulnerability in the pgAdmin 4 Maintenance Tool allows an authenticated user with tools_maintenance permissions to execute arbitrary SQL commands on the PostgreSQL server. Exploiting this vulnerability could lead to privilege escalation and execution of operating system commands on the database host.
- CVE-2026-17566Critical
A vulnerability in pgAdmin 4 allows remote code execution (RCE) via injection into the Import/Export Data tool. The validation function incorrectly handles backslashes in SQL strings, enabling bypass of protections and addition of a TO PROGRAM clause to psql commands.
- CVE-2026-17349Critical
In the adhoc_connect_server function in pgAdmin 4 9.0-9.16, when a non-owner user clones a server, all columns including credentials (passwords) and ownership flags are copied. This allows an attacker to take over the cloned server and use database passwords belonging to another user (e.g., an administrator).
- CVE-2026-12048Critical
A stored cross-site scripting vulnerability exists in pgAdmin 4 versions 6.0 through 9.15. Text returned by a PostgreSQL server (e.g., object names in errors or EXPLAIN fields) was passed verbatim to html-react-parser, allowing injection of arbitrary HTML, including iframes. An attacker controlling the server or creating a maliciously named object can hijack the victim's pgAdmin session.
- CVE-2026-12046Critical
In pgAdmin 4 versions 6.9 through 9.15, two SQL Editor endpoints lack authentication decorators, allowing unauthenticated attackers to trigger pickle deserialization from server sessions. Exploitation requires additional preconditions (knowledge of SECRET_KEY and write access to sessions directory) to achieve remote code execution.
- CVE-2026-12045Critical
A vulnerability in the pgAdmin 4 AI Assistant allows an attacker who can influence database content read by the assistant to execute arbitrary SQL with the privileges of the pgAdmin user's database role. The flaw stems from missing validation of LLM-generated queries, enabling bypass of read-only mode via injection of commands like COMMIT or ROLLBACK.
- CVE-2026-7813Critical
Authorization vulnerability in pgAdmin 4 server mode allows access to private objects of other users without proper filtering. An authenticated user could access another user's private servers, server groups, and background processes by guessing object IDs.
Original NVD description (English source)
The fix for CVE-2026-12045 in pgAdmin 4 9.16 required the LLM-supplied query passed to the AI Assistant's execute_sql_query tool to parse, via sqlparse, as exactly one non-transaction-control statement before running it inside a BEGIN TRANSACTION READ ONLY wrapper. sqlparse's string-literal lexing can disagree with PostgreSQL's own parser: under standard_conforming_strings = on (PostgreSQL's default since 9.1), a backslash immediately before a quote is an ordinary character to PostgreSQL, but sqlparse treats it as escaping the quote. A payload such as SELECT '\';COMMIT;CREATE TABLE pwn(x int);SELECT 1 --' therefore parses as a single SELECT to sqlparse's validator, while PostgreSQL executes it as four statements: the smuggled COMMIT ends the wrapping read-only transaction, and the trailing ROLLBACK becomes a no-op. This reintroduces the same write/RCE bypass CVE-2026-12045 was meant to close, reachable via the same indirect prompt-injection delivery (an attacker plants the payload in any object the AI Assistant may read; the LLM emits it as a tool call). An initial candidate fix ran the query with psycopg's execute(..., prepare=True), intending to force PostgreSQL's own Parse step (extended query protocol) to reject multi-statement text regardless of sqlparse's classification. This candidate fix does not work as submitted: psycopg3's PrepareManager silently ignores the prepare argument whenever the connection's prepare_threshold is None, which is pgAdmin's default for every server connection (the per-server "Prepare threshold" field is blank unless an administrator explicitly sets it) -- psycopg3 falls back to the simple query protocol, the same multi-statement-capable path the bypass exploits, so the candidate fix closes nothing on any real-world default configuration. The corrected fix sets conn.prepare_threshold = 0 directly on the dedicated, single-use read-only connection the AI Assistant tool opens, structurally forcing the extended query protocol independent of any server-level configuration. Verified against a live PostgreSQL 18 instance: the payload executes successfully under the prepare_threshold=None (default) behavior, and is rejected with "cannot insert multiple commands into a prepared statement" once prepare_threshold=0 is set on that connection. This issue affects pgAdmin 4: from 9.13 before 9.17.