Threats affecting pkg:github/pgAdmin 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 per-tool permission system (custom roles / role-based tool permissions, introduced in pgAdmin 4 9.3) did not enforce its permission check consistently. In SERVER mode, pgAdmin 4 gates each tool behind a per-tool Flask-Security permission, but the permission decorator (permissions_required) was applied only to a single "front door" route per tool. Every other backend route and Socket.IO handler in that tool's workflow relied solely on pga_login_required/socket_login_required, which check authentication but not the tool permission. The reporter verified three cases against a test build: (1) a user without tools_query_tool permission received 403 on the protected sqleditor initialization route, but the same session went on to connect the server, initialize the viewdata backend chain, and retrieve real table row content; (2) a user without tools_grant_wizard received 403 on the protected acl route, but the same session still enumerated grantable objects, generated GRANT SQL, and successfully applied it -- confirmed database-side via has_table_privilege(); (3) a user without tools_schema_diff received 403 on the protected panel route, but the same session initialized schema diff, enumerated and connected databases, and obtained real DDL differences via the compare_database Socket.IO handler. The reporter also confirmed a related but distinct issue: a non-owner triggering /misc/workspace/adhoc_connect_server against an administrator-owned shared server caused pgAdmin to persist a new server row still owned by the administrator (user_id/shared unchanged from the source), even though the connection attempt itself reported failure. During remediation, the same front-door-only permission gap was found to also affect the ERD, PSQL, and Debugger tools, and the Backup, Restore, Maintenance, and Import/Export blueprints, none of which were part of the original report; these were fixed using the same pattern as an extension of the reported defect class. An authenticated user who had valid pgAdmin login and a stored, working database connection, but had been explicitly denied a specific tool's permission by an administrator, could therefore still drive that tool end-to-end through its other routes and sockets, including obtaining an interactive psql session over the /pty Socket.IO namespace and invoking backup/restore/maintenance/import-export jobs. Because the bypass only restores access to tools operating over the user's own already-authenticated database connection, it does not grant the user any database privilege they did not already hold; it circumvents pgAdmin's own tool-level access-control policy (an organisational segregation-of-duties control, separate from database-level authorization), letting a user reach a pgAdmin feature an administrator intended to withhold from them, using capabilities their existing database role already permits through other means. Socket.IO event handlers had no permission-aware equivalent of permissions_required; only socket_login_required existed, checking authentication but not the tool permission. Fix adds a socket_permissions_required decorator (mirroring permissions_required, honouring the Administrator bypass, reading permissions via has_permission()) and applies it, alongside permissions_required, as the outermost decorator on every backend route and Socket.IO handler for the affected tools. Regression tests assert 403 on every gated route and socket handler for a permission-less user. This issue affects pgAdmin 4 in SERVER mode: from 9.3 before 9.17.

/misc/workspace/adhoc_connect_server, part of the Workspaces feature introduced in pgAdmin 4 9.0, when passed the id of an existing server, clones that server via Server.clone(), which copies every column from the source row, including user_id, shared, shared_username, and the stored credential fields password, save_password, and tunnel_password. When a non-owner triggered an adhoc connect against another user's (in practice, typically an administrator's) shared server, the clone inherited that user's ownership, shared flag, and stored database credentials verbatim. pgAdmin persisted this cross-tenant, credential-bearing server row before the connection was even attempted, so it survived even when the connection subsequently failed. The non-owner could then open the newly-owned clone and pgAdmin would connect using the source user's stored database password on the non-owner's behalf, granting the non-owner use of database credentials -- and whatever database privileges they confer -- that were never their own. Fix forces the cloned adhoc record's ownership fields (user_id, shared, shared_username) and stored credential fields (password, save_password, tunnel_password) to belong to the calling user and be cleared/private before committing, regardless of the source server's ownership, sharing state, or stored credentials. A regression test asserts that an adhoc connect triggered by a non-owner against another user's shared server persists a row owned by the caller, not shared, and without the source's stored credentials. This issue affects pgAdmin 4: from 9.0 before 9.17.

In SERVER mode, pgAdmin 4 enforces authentication per route via the @pga_login_required decorator; the application's before_request hook only handles desktop-mode auto-login and the Kerberos/Webserver-auth redirect, so any route shipped without the decorator is reachable without authentication (CWE-306). This is the same defect class previously fixed as CVE-2026-12046 (the sqleditor close/update_connection routes). A follow-up sweep, prompted by a report describing an incomplete fix for CVE-2026-12046, found further routes missing @pga_login_required: the Constraints blueprint's nodes and proplist (object listing) routes and its delete route (a state-mutating DELETE that removes table constraints); preferences.get_all_cli (GET, discloses all CLI-settable preference values); debugger.close (DELETE); and schema_diff.close (DELETE). An unauthenticated network client could therefore enumerate constraint metadata, delete table constraints, read preference values, and force-close debugger or schema-diff sessions belonging to other users, without ever authenticating. Fix adds the missing @pga_login_required decorator (and the corresponding import to the Constraints module) to each of these routes. The change is decorator-only; no behavioral changes to the underlying handlers. This issue affects pgAdmin 4 in SERVER mode: the Constraints and Debugger routes from 1.0, the Schema Diff close route from 4.18, and preferences.get_all_cli from 8.2, all 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.

The fix for CVE-2026-12044 in pgAdmin 4 9.16 hardened qtLiteral and switched sixteen COMMENT ON / pgstattuple / pgstatindex templates to it, but missed several sinks that had been placed in test_sql_string_literal_lint.py's ALLOWLIST on the incorrect assumption that schema, table, publication, and subscription names sourced from pg_catalog via the browser tree could never contain an apostrophe. PostgreSQL permits arbitrary characters in quoted identifiers, so a low-privileged user able to CREATE TABLE, CREATE PUBLICATION, or CREATE SUBSCRIPTION can plant an apostrophe'd object name that breaks out of the unescaped '{{ name }}' template interpolation the moment any user (including a higher-privileged one) opens that object's Statistics or Dependencies tab, allowing arbitrary SQL statement injection in the viewing user's database session. Affected sinks: the Index Statistics query for all-indexes listing (coll_stats.sql, both the 16_plus and default PostgreSQL-version template variants -- distinct from the single-index stats.sql path already fixed in CVE-2026-12044), and the publication and subscription dependencies.sql / get_position.sql templates (both the pg and ppas/EPAS dialect variants for publications). Fix switches all of these templates to qtLiteral(conn) for name interpolation, and updates publications/__init__.py and subscriptions/__init__.py to pass conn=self.conn into the dependencies.sql render_template call so the qtLiteral filter has a connection to quote against. The corresponding ALLOWLIST entries in test_sql_string_literal_lint.py are removed now that these sinks are properly escaped rather than merely assumed safe. A behavioral regression test renders each fixed template with a stacked-statement apostrophe payload and asserts both that the object name appears exactly as qtLiteral-escaped and that the rendered SQL parses as exactly one statement, verifying the assertion genuinely fails against the pre-patch raw-interpolation form. This issue affects pgAdmin 4: the Index Statistics sink from 1.0, and the Publications/Subscriptions sinks from 5.0, both before 9.17.

SQL injection in pgAdmin 4 across every dialog template that renders ``COMMENT ON ... IS '<description>'`` for a user-supplied description field. The Jinja templates for Domains (and their constraints), Foreign Tables, Languages, and Event Triggers, plus the Views OID-lookup query, interpolated the description directly inside a single-quoted SQL literal -- ``'{{ data.description }}'`` -- instead of passing it through the ``qtLiteral`` escape filter. An authenticated pgAdmin user with permission to create or alter the affected object types could submit a description containing an apostrophe, break out of the literal and chain arbitrary SQL. The injected SQL runs under the PostgreSQL role the user is already authenticated as; for a connected role with ``COPY ... TO/FROM PROGRAM`` (typically PostgreSQL superuser), this chains to OS command execution on the PostgreSQL host. The defect does not cross a privilege boundary -- the user already has direct SQL access to that role through pgAdmin's Query Tool -- so the attacker gains no capability beyond what their database role already grants. The marginal impact captures bypass of any application-layer Query Tool gating an operator may have configured. The defect was originally reported against the Domain Dialog ``description`` field; a code-wide audit identified sixteen sites of the same pattern across the templates listed above. The same review also surfaced ten related sinks in the pgstattuple/pgstatindex stats templates -- ``pgstattuple('{{schema}}.{{table}}')`` and the matching pgstatindex shape -- where ``qtIdent`` escapes embedded double quotes inside the identifier but not apostrophes, so a user with CREATE privilege on a schema could plant a table or index named ``foo'bar`` and a later stats viewer would render an unbalanced literal. Fix is layered: 1. Sites: replace every ``'{{ x.description }}'`` with ``{{ x.description|qtLiteral(conn) }}`` (no surrounding quotes -- the filter wraps the value in escaped quotes itself). Plumb ``conn=self.conn`` through every ``render_template`` call that loads one of these templates. Also corrects a ``{ % elif`` Jinja typo in the foreign-table schema diff (dead branch). Rewrite the ten pgstattuple/pgstatindex stats sites to address the relation via OID + ``::oid::regclass`` cast (e.g. ``pgstattuple({{ tid }}::oid::regclass)``), eliminating the embedded literal-call form entirely so that bug-class can no longer recur there. 2. Driver hardening: ``qtLiteral`` (in ``utils/driver/psycopg3/__init__.py``) used to silently return the raw unescaped value when its ``conn`` argument was falsy. It now raises ``ValueError`` -- surfacing the entire bug class going forward. The change immediately uncovered eight latent plumbing bugs (in ``schemas/__init__.py``, ``schemas/functions/__init__.py``, ``schemas/tables/utils.py``, ``foreign_servers/__init__.py``, and seven sites in ``roles/__init__.py``) -- all fixed as part of this patch. The inner ``except`` block that swallowed adapter-level failures and returned the raw value is also removed, so unadaptable inputs raise instead of leaking unescaped values. 3. Regression tests: a per-template behavioural test renders each previously-vulnerable template with an apostrophe-injection payload and asserts the escaped fragment is present and the vulnerable fragment absent; a lint test walks every ``*.sql`` template flagging any ``'{{ ... }}'`` single-quote-wrapped interpolation against an explicit allowlist; unit tests cover the new qtLiteral fail-fast and inner-except raise paths. This issue affects pgAdmin 4: from 1.0 before 9.16.

Open redirect in pgAdmin 4's multi-factor authentication flow. The MFA validate and register endpoints honoured the user-supplied 'next' query/form parameter without confirming the target pointed back inside pgAdmin, so an authenticated victim who clicked /mfa/validate?next=<external> -- a link typically delivered by phishing -- would be sent to an attacker-controlled host directly out of the trusted auth flow. The defect is a trusted-domain redirect, not a privilege bypass: the attacker gains no read/write access to pgAdmin or the victim's database, but the redirect launders the attacker's destination through pgAdmin's URL, which raises the success rate of credential-phishing follow-on against the victim. Fix introduces a same-origin _is_safe_redirect_url helper and gates every MFA redirect that consumes user-supplied 'next' values through it. The helper allows only relative paths and absolute URLs whose scheme is http(s) and whose host matches the current request host; it rejects external hosts in absolute and protocol-relative form, non-http schemes (javascript:, data:, mailto:), userinfo tricks (http://localhost@attacker/), and backslash variants that some browsers normalize to forward slashes. Unsafe targets fall back to the internal browser index. A dedicated regression test exercises each accept/reject category and the original reporter PoC. This issue affects pgAdmin 4: from 6.0 before 9.16.

Stored cross-site scripting in pgAdmin 4's error-rendering and plan-node-rendering paths. Text returned by a PostgreSQL server (ErrorResponse messages, including object names quoted back inside relation-does-not-exist errors and inside EXPLAIN Recheck Cond / Exact Heap Blocks fields) was passed verbatim through html-react-parser at every user-facing sink — the notifier toasts, FormFooterMessage / FormInput help and error areas, FormNote, ModalProvider AlertContent and confirmDelete, ToolErrorView, the Explain visualiser's NodeText panel, the SQL editor confirm dialogs, ConfirmSaveContent, PreferencesHelper modal alerts, and SelectThemes helper text. A PostgreSQL server an attacker controls — or any server returning attacker-influenced text such as a table or column name a low-privilege database user can create — could inject arbitrary HTML (including <iframe>) into the pgAdmin DOM the moment the victim's pgAdmin connected to that server or viewed an Explain plan that referenced the crafted object. The injected iframe's srcdoc could fetch attacker-served JavaScript and, by writing to parent.location, redirect the victim's top-level pgAdmin browser tab to an attacker-controlled URL. Because the injection originates from inside pgAdmin's own interface, standard anti-clickjacking controls (X-Frame-Options, Content-Security-Policy: frame-ancestors) do not mitigate it. A phishing page rendered inside the legitimate pgAdmin window is indistinguishable from a genuine pgAdmin dialog. Fix combines three complementary layers. (1) DOMPurify sanitisation is wrapped around every html-react-parser call site reachable from notifier, alert, form-error, Explain, and SQL-editor flows. (2) A new plain-text rendering contract — SafeMessage / SafeHtmlMessage components plus Notifier.errorText / alertText / warningText / infoText / successText helpers — is introduced; around fifty callers across browser, tools, dashboard, debugger, misc, llm, preferences, schema diff, and the SQL editor that previously interpolated backend-derived strings are migrated to the plain-text variants. (3) Backend HTML-escape is applied at the post-connection-SQL handler (execute_post_connection_sql) via a new sanitize_external_text helper, so third-party JSON consumers (audit logs, API clients) never receive raw markup either; the Explain plan-info renderer is also patched to _.escape Recheck Cond and Exact Heap Blocks at construction (matching every sibling field), giving defence in depth even before DOMPurify runs. This issue affects pgAdmin 4: from 6.0 before 9.16.

HTML injection in pgAdmin 4's cloud deployment module. The verify_credentials, deploy, regions, and update-server endpoints under /rds/, /azure/, /google/, and the top-level /cloud/ blueprint propagated AWS / Azure / Google SDK exception text — and the related file-resolution and database-commit exception text — into the JSON response body (the info and errormsg fields) without HTML-encoding. The Cloud Wizard frontend rendered these strings through html-react-parser, so an attacker-influenced exception message embedded structural HTML directly into the wizard's DOM. The reported entry point is /rds/verify_credentials/. An authenticated pgAdmin user submits a crafted access_key whose value contains an <iframe/src=...> payload; AWS STS rejects the credential with an IncompleteSignature exception whose text quotes the access_key verbatim; the pgAdmin backend forwards that text into the JSON info field; the Cloud Wizard's FormFooterMessage parses it as HTML. The browser fetches the iframe's src from an attacker-controlled host, and JavaScript executing inside the cross-origin iframe writes to parent.location, redirecting the victim's pgAdmin tab. Because the injection renders inside pgAdmin's own interface, X-Frame-Options and Content-Security-Policy frame-ancestors do not mitigate it. Baseline impact is self-targeted (the same user who supplied the payload sees the injection); escalation against other authenticated users requires an additional cross-site request-forgery primitive capable of submitting the malformed credential request with a valid X-pgA-CSRFToken in the victim's browser context. The same unsanitised-error-into-JSON pattern was present across multiple sibling endpoints — Azure's check_cluster_name_availability, every Google endpoint that surfaces SDK errors (verification_ack, projects, regions, instance_types, database_versions, the verify_credentials path-resolution branches), the central /deploy endpoint that bubbles str(e) from deploy_on_rds / deploy_on_azure / deploy_on_google, and update_cloud_server which surfaces the str(e) from a failing db.session.commit — all of which are now covered. Fix HTML-escapes every external/SDK exception string at the endpoint sink via a new shared sanitize_external_text helper (HTML escape with control-character strip), promoted out of the psycopg3 driver into web/pgadmin/utils/text_sanitize.py. The Cloud Wizard frontend additionally renders its FormFooterMessage in plain-text mode for backend-derived strings, so the value is never parsed as HTML even if a future sink forgets the escape. This issue affects pgAdmin 4: from 6.6 before 9.16.

Two state-mutating endpoints in pgAdmin 4's SQL Editor blueprint -- DELETE /sqleditor/close/<trans_id> and POST /sqleditor/initialize/sqleditor/update_connection/<sgid>/<sid>/<did> -- were the only routes in the module missing the @pga_login_required decorator. Both reach a pickle.loads sink on session['gridData'][<trans_id>]['command_obj']: the close endpoint via close_sqleditor_session(), and update_sqleditor_connection via check_transaction_status(). In server mode these endpoints were reachable without any authenticated pgAdmin session. The defect is a missing-authentication-on-critical-function (CWE-306) wrapper around a deserialization-of-untrusted-data sink (CWE-502). Exploiting it for remote code execution requires the attacker to also forge a server-side session file whose gridData entry contains a malicious pickle payload, which in turn requires both (a) knowledge of pgAdmin's Flask SECRET_KEY (no chain to leak it is described here -- the attacker must already possess it) and (b) write access to pgAdmin's sessions/ directory on the host. Neither precondition is granted by this defect on its own. When those preconditions are met from another channel (misconfigured deployment, prior compromise, leaked configuration), the missing auth gate is the final hop that turns an existing partial compromise into unauthenticated code execution in the pgAdmin process -- and, by extension, on the host under whatever account runs pgAdmin. Fix is a one-line @pga_login_required decorator on each of the two endpoints, matching the convention used by every other route in the module. The is_authenticated / MFA chain now runs before the trans_id is dereferenced, so an unauthenticated request is rejected before reaching the deserialization path. The defect is server-mode only. In DESKTOP mode pgAdmin's before_request hook re-authenticates DESKTOP_USER on every request, so no endpoint can be exercised in an unauthenticated state and no auth decorator (or its absence) is meaningful. The accompanying regression test mirrors the attacker's path -- harvests an X-pgA-CSRFToken from GET /login and replays it against both endpoints -- and self-skips outside server mode for that reason; it is wired into the existing server-mode CI workflow alongside the data-isolation tests. This issue affects pgAdmin 4: from 6.9 before 9.16.

Read-only transaction bypass in the pgAdmin 4 AI Assistant allows an attacker who can influence database content that the assistant reads to execute arbitrary SQL with the privileges of the pgAdmin user's database role. The AI Assistant's execute_sql_query tool runs LLM-generated SQL inside a BEGIN TRANSACTION READ ONLY wrapper to prevent data modification. The LLM-supplied query was forwarded to the database driver without restriction to a single statement or to read-only verbs, so a multi-statement payload beginning with COMMIT, END, ROLLBACK, or ABORT terminated the read-only transaction and ran subsequent statements in autocommit mode. The trailing ROLLBACK then had no effect. Delivery is via prompt injection: an attacker who can write content into any object the AI Assistant may inspect (a row, a column value, a comment) can cause the LLM to emit the multi-statement payload as a tool call. With ordinary write privileges on the pgAdmin user's role the attacker can perform unauthorised data modification. When the pgAdmin user's role is a PostgreSQL superuser or holds pg_execute_server_program, the chain extends to remote code execution on the database server host via COPY ... TO PROGRAM. Fix validates the LLM-supplied query up front: it must parse to exactly one non-empty / non-comment statement whose leading real token (after stripping whitespace, comments, and punctuation) is one of SELECT, WITH, EXPLAIN, SHOW, VALUES, or TABLE. Transaction-control verbs, DML, DDL, CALL, COPY, DO, SET/RESET, and everything else are rejected before any database work happens. PostgreSQL's READ ONLY mode continues to backstop data-modifying CTEs, EXPLAIN ANALYZE on writes, and volatile side effects. This issue affects pgAdmin 4: from 9.13 before 9.16.

SQL injection in pgAdmin 4's named restore point endpoint (POST /browser/server/restore_point/{gid}/{sid}). The user-supplied 'value' field was interpolated directly into the SQL string with str.format() instead of being passed as a bound parameter, allowing an authenticated pgAdmin user with a connected PostgreSQL session to inject additional statements through that endpoint. The injected SQL executes under the database role the user is already authenticated as. The defect does not cross a privilege boundary -- the user already has direct SQL access to that role through the Query Tool -- so the attacker gains no capability beyond what their database role already grants them. The marginal impact accounts for the fact that the injection path is not the documented SQL-execution interface, so a deployment that gates the Query Tool at the application layer could see SQL executed through a path it did not anticipate. Fix passes the restore point name as a bound parameter and schema-qualifies the function call as pg_catalog.pg_create_restore_point so a non-default search_path on the connection cannot redirect the call to a shadow definition. A regression test asserts the value arrives as a bound parameter and not spliced into the SQL string. This issue affects pgAdmin 4: from 1.0 before 9.16.

Improper restriction of excessive authentication attempts (CWE-307) in pgAdmin 4.pgAdmin enforces MAX_LOGIN_ATTEMPTS only inside its custom /authenticate/login view. Flask-Security's default /login view, which is registered automatically by security.init_app() and is reachable on every server, never consulted the User.locked field: pgAdmin's User model relied on Flask-Security's UserMixin.is_locked() (which always returns 'not locked') and Flask-Login's is_active (which only checks the active column, not locked). An attacker who triggered an account lockout via /authenticate/login could therefore obtain a session by re-submitting valid credentials directly to /login, defeating the brute-force-protection control for accounts using the INTERNAL authentication source. The same bypass also means that login attempts via /login are never rate-limited, so an attacker can perform an unbounded online password-guessing attack against INTERNAL accounts regardless of MAX_LOGIN_ATTEMPTS. Fix overrides User.is_active and User.is_locked() so the locked column is enforced on every authentication path. LDAP, OAuth2, Kerberos, and Webserver users are not reachable by this bypass because they have no local password and are rejected by Flask-Security's LoginForm.validate before the locked check; the lockout itself is also internal-only (the /authenticate/login view filters by auth_source=INTERNAL). This issue affects pgAdmin 4: before 9.15.

Symbolic-link path traversal (CWE-61, CWE-22) in pgAdmin 4 File Manager.check_access_permission used os.path.abspath, which resolves '..' but does not resolve symbolic links, while the subsequent kernel write follows symlinks. An authenticated user could plant a symbolic link inside their own storage directory pointing outside it and induce pgAdmin to write to any path reachable by the pgAdmin process. Fix switches the access check to os.path.realpath for both source and destination, and adds an _open_upload_target helper that opens the target with O_NOFOLLOW (mode 0o600) to close the leaf-component TOCTOU between the access check and the open. File mode is hardened from 0o644 to 0o600. This issue affects pgAdmin 4: before 9.15.

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.

Local file inclusion (LFI) and server-side request forgery (SSRF) vulnerabilities in pgAdmin 4 LLM API configuration endpoints. User-supplied api_key_file and api_url preferences were passed to the LLM provider clients without validation. An authenticated user could read arbitrary server-side files by pointing api_key_file at any path readable by the pgAdmin process, or coerce pgAdmin into making requests to internal targets (e.g.cloud metadata services such as 169.254.169.254) by setting api_url, exploiting the chat path and model-list endpoints. Fix restricts api_key_file to the user's private storage (server mode) or home directory (desktop mode), enforces a printable-ASCII key shape and a 1024-byte read cap, and gates api_url against a configurable allow-list (config.ALLOWED_LLM_API_URLS) at every entry point. This issue affects pgAdmin 4: before 9.15.

OS command injection (CWE-78) vulnerability in pgAdmin 4 Import/Export query export. User-supplied input was interpolated directly into a psql \copy metacommand template without sanitization. An authenticated user could inject ") TO PROGRAM 'cmd'" to break out of the \copy (...) context and achieve arbitrary command execution on the pgAdmin server, or ") TO '/path'" for arbitrary file write. Additional fields (format, on_error, log_verbosity) were also raw-interpolated and exploitable. Fix adds a parens-balance parser modeled on psql's strtokx tokenizer, allow-lists format/on_error/log_verbosity, rejects null bytes in the query, and tightens type and gating checks. This issue affects pgAdmin 4: before 9.15.

SQL injection vulnerability in pgAdmin 4 Maintenance Tool. Four user-supplied JSON fields (buffer_usage_limit, vacuum_parallel, vacuum_index_cleanup, reindex_tablespace) were concatenated directly into the rendered VACUUM/ANALYZE/REINDEX command and passed to psql --command. An authenticated user with the tools_maintenance permission could break out of the option syntax and execute arbitrary SQL on the connected PostgreSQL server. The injected SQL could in turn invoke COPY ... TO PROGRAM to escalate to operating-system command execution on the database host. Fix introduces server-side allow-listing of all four fields and switches reindex_tablespace from manual quoting to the qtIdent filter. This issue affects pgAdmin 4: before 9.15.

Stored cross-site scripting (XSS) vulnerability in pgAdmin 4 Browser Tree and Explain Visualizer modules. User-controlled PostgreSQL object names (database, schema, table, column, etc.) were assigned to DOM elements via innerHTML, allowing crafted object names containing HTML markup to execute attacker-supplied JavaScript in the browser of any pgAdmin user who navigated to or executed EXPLAIN over the malicious object. Fix replaces innerHTML with textContent. This issue affects pgAdmin 4: before 9.15.