| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| 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. |
| pgAdmin 4's Import/Export Data tool builds a psql \copy (...) command line by interpolating a user-supplied SQL query into a Jinja template and passing the rendered line to psql via --command. To stop an attacker from breaking out of the (...) wrapper, create_import_export_job() (route POST /import_export/job/<sid>, gated only by the ordinary, commonly-granted tools_import_export_data permission) validated the query with a hand-written parenthesis-balance checker, _is_query_parens_balanced(). That checker always treated a backslash before a single quote (\') as escaping the quote, i.e. as if standard_conforming_strings were off. PostgreSQL has defaulted standard_conforming_strings to on since 9.1 (2010), the default on every PostgreSQL version pgAdmin 4 currently supports (13-18); under that default psql's own \copy tokenizer treats \ as an ordinary character, so a single quote immediately after it closes the string literal. A query such as SELECT 'a\') TO PROGRAM 'echo pwned' x' was therefore accepted as "balanced" by pgAdmin's checker (which believed the ) was still inside the string), while psql, run through the actual rendered command line, closes the string at that point and treats the following ) as the end of the wrapping \copy (...) subquery, exposing an attacker-chosen TO PROGRAM '<command>' clause that psql executes via popen() -- independent of a subsequent syntax error later on the same line. This is the same class of bug as CVE-2025-12762/CVE-2025-13780 (RCE via psql meta-command/COPY injection during PLAIN-format dump restore), reached through an independently written defense in a different module (Import/Export Data rather than Restore) that had its own, different logic bug (inverted backslash-escape semantics rather than a BOM-defeated regex anchor).
The fix rejects any backslash inside a single-quoted string in the query outright, rather than picking one of the two possible psql interpretations. This is intentionally conservative: because the correct interpretation of \ depends on the target server's standard_conforming_strings setting, which the checker cannot reliably know at validation time, refusing the query is safer than guessing.
This issue affects pgAdmin 4: from the introduction of _is_query_parens_balanced() before 9.18. |
| A flaw was found in libsoup. The HTTP/2 server in libsoup may not fully validate the values of pseudo-headers :scheme, :authority, and :path, which may allow a user to cause a denial of service (DoS). |
| An attacker can send a crafted EDNS OPT record that will be ignored by DNSdist’s filtering rules, but will be rewritten as a valid OPT record when EDNS Client Subnet is inserted, causing the backend to see the EDNS option(s) that DNSdist did not filter. |
| node-tar is a full-featured Tar for Node.js. Prior to 7.5.16, tar (node-tar) applies a PAX extended header's size= record (and other PAX overrides) to the next header entry of any type, including intermediary metadata headers such as a GNU long-name (L) or long-link (K) entry. Per POSIX pax, a PAX extended header (x) describes the next file entry, not the intermediary extension headers that may sit between the x header and the file it annotates. Because node-tar lets the PAX size override the byte length of an intervening L/K/x header, an attacker can desynchronize node-tar's stream cursor relative to every other mainstream tar implementation (GNU tar, libarchive/bsdtar, Python tarfile, and the now-fixed tar-rs / astral-tokio-tar). The result is a tar parser interpretation differential (CWE-436): a single crafted archive yields a different set of members under node-tar than under the reference tar tools. An attacker can use this to hide a member from one parser while it is visible to another, which defeats security tooling whose scanner and extractor disagree on archive contents (e.g. a malware/secret scanner that lists entries with one library while a downstream step extracts with another) This vulnerability is fixed in 7.5.16. |
| A flaw was found in vLLM, an open-source library for large language model inference. This vulnerability arises from improper handling of image metadata, specifically EXIF orientation and PNG transparency (tRNS) data, during image processing. When images are converted to RGB, transparency information may be implicitly discarded or remapped, leading to unexpected rendering of transparent pixels and distortion of input content. This can result in the model misinterpreting image content, potentially affecting the integrity of processed data. |
| In the Linux kernel, the following vulnerability has been resolved:
netfilter: nf_tables: reject immediate NF_QUEUE verdict
nft_queue is always used from userspace nftables to deliver the NF_QUEUE
verdict. Immediately emitting an NF_QUEUE verdict is never used by the
userspace nft tools, so reject immediate NF_QUEUE verdicts.
The arp family does not provide queue support, but such an immediate
verdict is still reachable. Globally reject NF_QUEUE immediate verdicts
to address this issue. |
| Spring MVC and WebFlux applications are vulnerable to stream corruption when using Server-Sent Events (SSE). This issue affects Spring Foundation: from 7.0.0 through 7.0.5, from 6.2.0 through 6.2.16, from 6.1.0 through 6.1.25, from 5.3.0 through 5.3.46. |
| Rack is a modular Ruby web server interface. From versions 3.0.0.beta1 to before 3.1.21 and 3.2.0 to before 3.2.6, Rack::Utils.forwarded_values parses the RFC 7239 Forwarded header by splitting on semicolons before handling quoted-string values. Because quoted values may legally contain semicolons, a header can be interpreted by Rack as multiple Forwarded directives rather than as a single quoted for value. In deployments where an upstream proxy, WAF, or intermediary validates or preserves quoted Forwarded values differently, this discrepancy can allow an attacker to smuggle host, proto, for, or by parameters through a single header value. This issue has been patched in versions 3.1.21 and 3.2.6. |
| Matching of hosts against proxy patterns can improperly treat an IPv6 zone ID as a hostname component. For example, when the NO_PROXY environment variable is set to "*.example.com", a request to "[::1%25.example.com]:80` will incorrectly match and not be proxied. |
| The ParseAddressList function incorrectly handles comments (text within parentheses) within display names. Since this is a misalignment with conforming address parsers, it can result in different trust decisions being made by programs using different parsers. |
| ALTCHA is privacy-first software for captcha and bot protection. A cryptographic semantic binding flaw in ALTCHA libraries allows challenge payload splicing, which may enable replay attacks. The HMAC signature does not unambiguously bind challenge parameters to the nonce, allowing an attacker to reinterpret a valid proof-of-work submission with a modified expiration value. This may allow previously solved challenges to be reused beyond their intended lifetime, depending on server-side replay handling and deployment assumptions. The vulnerability primarily impacts abuse-prevention mechanisms such as rate limiting and bot mitigation. It does not directly affect data confidentiality or integrity. This issue has been addressed by enforcing explicit semantic separation between challenge parameters and the nonce during HMAC computation. Users are advised to upgrade to patched versions, which include version 1.0.0 of the altcha Golang package, version 1.0.0 of the altcha Rubygem, version 1.0.0 of the altcha pip package, version 1.0.0 of the altcha Erlang package, version 1.4.1 of the altcha-lib npm package, version 1.3.1 of the altcha-org/altcha Composer package, and version 1.3.0 of the org.altcha:altcha Maven package. As a mitigation, implementations may append a delimiter to the end of the `salt` value prior to HMAC computation (for example, `<salt>?expires=<time>&`). This prevents ambiguity between parameters and the nonce and is backward-compatible with existing implementations, as the delimiter is treated as a standard URL parameter separator. |
| A firmware bug which may lead to misinterpretation of data in the AMC2-4WCF and AMC2-2WCF allowing an adversary to grant access to the last authorized user. |
| Misinterpretation of Input vulnerability in OpenText™ Service Management Automation X (SMAX), OpenText™ Asset Management X (AMX), and OpenText™ Hybrid Cloud Management X (HCMX) products. The vulnerability could allow Input data manipulation.This issue affects Service Management Automation X (SMAX) versions: 2020.05, 2020.08, 2020.11, 2021.02, 2021.05, 2021.08, 2021.11, 2022.05, 2022.11, 2023.05; Asset Management X (AMX) versions: 2021.08, 2021.11, 2022.05, 2022.11, 2023.05; and Hybrid Cloud Management X (HCMX) versions: 2020.05, 2020.08, 2020.11, 2021.02, 2021.05, 2021.08, 2021.11, 2022.05, 2022.11, 2023.05.
|
| OpenSSH before 10.3 can use unintended ECDSA algorithms. Listing of any ECDSA algorithm in PubkeyAcceptedAlgorithms or HostbasedAcceptedAlgorithms is misinterpreted to mean all ECDSA algorithms. |
| crun is an open source OCI Container Runtime fully written in C. In versions 1.19 through 1.26, the `crun exec` option `-u` (`--user`) is incorrectly parsed. The value `1` is interpreted as UID 0 and GID 0 when it should have been UID 1 and GID 0. The process thus runs with higher privileges than expected. Version 1.27 patches the issue. |
| Astro is a web framework for content-driven websites. In versions of astro before 5.13.2 and 4.16.18, the image optimization endpoint in projects deployed with on-demand rendering allows images from unauthorized third-party domains to be served. On-demand rendered sites built with Astro include an /_image endpoint which returns optimized versions of images. A bug in impacted versions of astro allows an attacker to bypass the third-party domain restrictions by using a protocol-relative URL as the image source, e.g. /_image?href=//example.com/image.png. This vulnerability is fixed in 5.13.2 and 4.16.18. |
| Autel MaxiCharger AC Wallbox Commercial ble_process_esp32_msg Misinterpretation of Input Vulnerability. This vulnerability allows network-adjacent attackers to inject arbitrary AT commands on affected installations of Autel MaxiCharger AC Wallbox Commercial charging stations. Authentication is not required to exploit this vulnerability.
The specific flaw exists within the ble_process_esp32_msg function. The issue results from misinterpretation of input data. An attacker can leverage this vulnerability to execute AT commands in the context of the device. Was ZDI-CAN-26368. |
| WOLFBOX Level 2 EV Charger MCU Command Parsing Misinterpretation of Input Remote Code Execution Vulnerability. This vulnerability allows network-adjacent attackers to execute arbitrary code on affected installatons of WOLFBOX Level 2 EV Charger devices. Authentication is required to exploit this vulnerability.
The specific flaw exists within the handling of command frames received by the MCU. When parsing frames, the process does not properly detect the start of a frame, which can lead to misinterpretation of input. An attacker can leverage this in conjunction with other vulnerabilities to execute arbitrary code in the context of the device. Was ZDI-CAN-26501. |
| GitProxy is an application that stands between developers and a Git remote endpoint (e.g., github.com). In versions 1.19.1 and below, an attacker can craft a malicious Git packfile to exploit the PACK signature detection in the parsePush.ts file. By embedding a misleading PACK signature within commit content and carefully constructing the packet structure, the attacker can trick the parser into treating invalid or unintended data as the packfile. Potentially, this would allow bypassing approval or hiding commits. This issue is fixed in version 1.19.2. |
