CVE-2026-12045: Pgadmin 4 Command Injection (CVSS 9.4)

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.

The MASTER_PASSWORD_HOOK setting, introduced in pgAdmin 4 7.2, lets an administrator configure an external command that returns a per-user encryption key, with %u in the configured string replaced by the current user's name. The previous implementation substituted the username directly into the command string and executed the result with subprocess.Popen(..., shell=True). Because the username can originate from an external authentication source (OAuth/OIDC claims, Kerberos, webserver auth) rather than a value pgAdmin fully controls, a username containing shell metacharacters (';', '$()', backticks, pipes, '&&', newlines) allowed an authenticated user to execute arbitrary commands as the pgAdmin service account in any deployment where the configured hook string uses %u. Fix tokenises the trusted, administrator-configured hook string into an argument vector first (using shlex in POSIX-quoting mode, with backslash-escaping disabled so Windows-style paths are not mis-parsed), substitutes the untrusted username into the individual argv elements, and executes with shell=False. The username is therefore always confined to a single argv element; any shell metacharacters it contains are inert. Administrators whose MASTER_PASSWORD_HOOK previously relied on shell features (pipes, redirection, environment-variable expansion, globbing) within the hook string itself must move that logic into the invoked script, since it is no longer interpreted by a shell. This issue affects pgAdmin 4: from 7.2 before 9.17.

Deserialization of untrusted data (CWE-502) in pgAdmin 4 FileBackedSessionManager. The session manager performed unsafe deserialization of session-file contents (using Python's standard object-serialization module) before performing any HMAC integrity check. Any file dropped into the sessions directory was deserialized unconditionally. An authenticated user with write access to the sessions directory (whether by misconfiguration or in combination with another path-traversal flaw) could plant a crafted serialized payload to achieve operating-system level remote code execution under the pgAdmin process identity. Fix prepends a 64-byte hex SHA-256 HMAC over the session body, computed with SECRET_KEY, and verifies it via hmac.compare_digest before any deserialization. The check is raised (rather than asserted) on empty SECRET_KEY so it is not stripped under -O. This issue affects pgAdmin 4: before 9.15.

Crafted MQTT messages can trigger command injection, resulting in root-level code execution on the target device.

A command injection vulnerability exists in MLflow's model serving container initialization code, specifically in the `_install_model_dependencies_to_env()` function. When deploying a model with `env_manager=LOCAL`, MLflow reads dependency specifications from the model artifact's `python_env.yaml` file and directly interpolates them into a shell command without sanitization. This allows an attacker to supply a malicious model artifact and achieve arbitrary command execution on systems that deploy the model. The vulnerability affects versions 3.8.0 and is fixed in version 3.8.2.

This vulnerability allows authenticated attackers to execute arbitrary commands on the underlying system using the file name of an uploaded file.