| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| mistral.rs before 0.9.5 contains a delimiter injection vulnerability in mistralrs-core that allows attackers to inject tool calls by placing delimiter strings in text the model quotes. Attackers controlling documents, web pages or tool outputs can embed tags like </tool_call> to make qwen-xml, gemma4 or reasoning parsers return extra calls to declared tools with attacker-chosen arguments. |
| Lavish is the new editor for your HTML artifacts. Prior to 0.1.43, the lavish-axi loopback HTTP server did not authenticate requests or validate Host and Origin values on session, artifact, prompt, and polling routes. A website visited while the server is running can use DNS rebinding to send same-origin browser requests to the local service. POST /api/sessions accepts a caller-selected file path and returns its deterministic session key; GET /artifact/:key/index.html then returns the file contents, allowing disclosure of any file readable by the server process without an existing session. An attacker can also queue text through POST /api/:key/prompts for an active agent session when the artifact path is known or guessed. Visiting the attacker-controlled website is the required user interaction. This vulnerability is fixed in 0.1.43. |
| CallBackery versions before 0.62.0 for Perl accept session cookies of any age sent in the X-Session-Cookie header.
cookieConf() rejects a cookie older than 300 seconds only when it arrives in the xsc query parameter. The same cookie sent in the X-Session-Cookie header is accepted however old it is.
A session cookie that leaks through a log, a Referer header or browser history authenticates as its user indefinitely. |
| Red Hat Product Security has been notified that this CVE is a duplicated of CVE-2026-104047 |
| Improper neutralization of input used for LLM prompting vulnerability in Johnson Controls EasyIO Neo IO Manager and Tools Installer allows Search Order Hijacking.
This issue affects EasyIO Neo IO Manager and Tools Installer: 1.1.0.0. |
| A security vulnerability has been detected in esengine DeepSeek-Reasonix up to 2.52.0. Affected is the function discoverDocs of the file internal/memory/doc.go. The manipulation leads to improper neutralization of input used for llm prompting. The attack can be initiated remotely. The exploit has been disclosed publicly and may be used. Upgrading to version 2.53.0 is able to address this issue. The identifier of the patch is 4ca33493a673c2788aa8598fbca2b2f47a69e1b8. It is suggested to upgrade the affected component. |
| In JetBrains Datalore before 2026.4 account takeover was possible via SSO sign-in to a pre-registered account |
| In JetBrains Datalore before 2026.4 aPI tokens and sessions of removed or banned users remained valid |
| OpenHuman is an open-source agent harness available as a desktop application and Rust library. Before 0.57.13, background synchronization for connected services such as email, Slack, and Notion stores raw external content and later places it in the model context of autonomous background agents without neutralizing embedded instructions. ApprovalGate treats tool calls without an interactive chat context as pre-authorized automation, so attacker-authored content processed during a background tick can influence the model to invoke Composio actions without presenting an approval request. An attacker who can deliver content to a connected account can use this path to send or forward data, post messages, create or modify third-party content, or disclose information from OpenHuman memory and integrations. The issue requires an active integration and successful model interpretation of the injected instructions. This issue is fixed in version 0.57.13. |
| Sogou Input Method for Windows contains a vulnerability in the configuration retrieval mechanism of its SGTool.exe component. The application retrieves configuration data over unencrypted HTTP connections, allowing an attacker capable of manipulating network traffic to replace legitimate server responses with malicious configurations. An attacker can exploit this vulnerability to instruct SGTool.exe to download attacker-controlled files to arbitrary locations, including paths used by legitimate application components. This can result in arbitrary code execution when the downloaded files are subsequently executed or loaded by the application. |
| OpenHuman is an open-source agent harness available as a desktop application and Rust library. Before 0.57.13, OpenHuman places names, descriptions, titles, and parameter definitions returned by configured external MCP servers into the agent model context without neutralizing instructions or bounding the descriptive fields. The McpRemoteTool records produced by list_tools preserve this remote content, while tool filtering considers the tool name and the prompt-injection detector screens user messages rather than MCP definitions. A malicious or compromised MCP server can therefore return a poisoned description that is evaluated on every agent turn and may direct the model to invoke other available tools, disclose accessible data, or perform unauthorized actions without a separate malicious user message. The issue affects users who configure an attacker-controlled or compromised MCP server. This issue is fixed in version 0.57.13. |
| Improper neutralization of input used for LLM prompting in the environment tool in Strands Agents Tools before 0.8.6 might allow remote actors to obtain the agent host's environment variables in cleartext, including cloud credentials and API keys, via a crafted prompt that disables the tool's secret masking.
To remediate this issue, users should upgrade to version 0.8.6 or later and rotate any credentials that were present in the environment of agents that loaded the environment tool. |
| A flaw was found in SSSD. When configured to use Microsoft Entra ID, search inputs are not properly sanitized before being incorporated into directory query filters. A local user can exploit this vulnerability by submitting a crafted lookup request, manipulating the query logic to cause unauthorized information disclosure from the directory. |
| A flaw was found in Quarkus where HTTP security policies are not sanitizing certain character permutations correctly when accepting requests, resulting in incorrect evaluation of permissions. This issue could allow an attacker to bypass the security policy altogether, resulting in unauthorized endpoint access and possibly a denial of service. |
| ### Summary
`qs.stringify` with `allowEmptyArrays: true` writes the key of an empty array without encoding it, even though keys are encoded by default. A key that contains `&` (or the configured `delimiter`), `=`, or `#` therefore becomes query string syntax instead of data, so a caller-supplied key name can add parameters to the serialized query string, or cut it short.
### Details
In `lib/stringify.js`, the `allowEmptyArrays` branch returns `adjustedPrefix + '[]'` early, before the code that passes every other key through `encoder` and `formatter`. So `{ "x&k=v": ["1"] }` serializes as `x%26k%3Dv%5B0%5D=1`, but `{ "x&k=v": [] }` serializes as `x&k=v[]`. The early return applies at any depth (`{ "a": { "b&k=v": [] } }` gives `a[b&k=v][]`), under every `arrayFormat`, with `allowDots`, and with a custom `encoder`, which is never called for that key. With `encode: false` or `encodeValuesOnly: true`, keys are intentionally left unencoded on every path; those configurations are not affected.
#### PoC
```js
var qs = require("qs");
var query = qs.stringify({ q: "hi", "x&apiKey=evil&y": [] }, { allowEmptyArrays: true });
// "q=hi&x&apiKey=evil&y[]"
qs.parse(query);
// { q: "hi", x: "", apiKey: "evil", y: [""] }
qs.stringify({ "x#": [], z: "1" }, { allowEmptyArrays: true });
// "x#[]&z=1": in a URL, everything after "#" becomes the fragment
```
#### Fix
Applied in 234d505 on `main` and released as v6.17.0, and backported to v6.16.2, v6.15.4, v6.14.3, v6.13.4, and v6.12.6: the key of an empty array goes through `encoder` and `formatter` exactly as the key of any other value does, so `{ "x&apiKey=evil&y": [] }` now serializes as `x%26apiKey%3Devil%26y[]`. Keys that need no encoding serialize as before, with two exceptions that match every other path: the brackets of a nested key are now encoded (`a[b][]` becomes `a%5Bb%5D[]`, which `qs.parse` reads identically), and with `encodeDotInKeys`, a dot in the key becomes `%252E` instead of `%2E` (which `decodeDotInKeys` previously misread as nesting).
### Affected versions
- `>=6.12.0 <6.12.6`, fixed in v6.12.6
- `>=6.13.0 <6.13.4`, fixed in v6.13.4
- `>=6.14.0 <6.14.3`, fixed in v6.14.3
- `>=6.15.0 <6.15.4`, fixed in v6.15.4
- `>=6.16.0 <6.16.2`, fixed in v6.16.2
v6.17.0 and later are not affected.
`allowEmptyArrays` was introduced in f22b2bc and first shipped in v6.12.0, with this early return. Earlier versions have no `allowEmptyArrays` option and are not affected.
### Impact
An application that passes an object whose key names come from untrusted input to `qs.stringify` with `allowEmptyArrays: true` (for example, to forward a parsed JSON body or user-selected fields to an upstream service) can have parameters injected into the resulting query string. The injected parameters are not keys of the object, so deleting or overwriting a key on the object (such as setting a trusted `apiKey` after copying the user's fields) does not remove them; whether the attacker's value or the application's wins depends on how the receiving side handles a duplicate parameter. A `#` in such a key ends the query string early when the result is placed in a URL. `allowEmptyArrays` is off by default, and only keys whose value (after any `filter`) is an empty array are affected. |
| CGI::Cookie versions before 5.00 for Perl allow cookie attribute injection via unvalidated path, domain and expires values.
The path, domain and expires values are written into the Set-Cookie header without checking for semicolons or control characters.
An application that sets them from request data can be made to add other cookie attributes, such as Domain or SameSite. |
| CGI versions before 5.00 for Perl allow Content-Disposition parameter injection via unescaped double quotes in the header attachment filename.
header(-attachment => $name) writes $name into the quoted filename parameter of Content-Disposition without escaping double quotes.
An application that passes request data as the attachment name can be made to send a second filename parameter, changing the name the response is saved under. |
| A vulnerability in the CLI of Cisco IOS XE Software could allow an authenticated, local attacker with administrative privileges to execute arbitrary commands as root on the underlying operating system of an affected device.
This vulnerability is due to insufficient validation of user arguments that are passed to specific CLI commands. An attacker could exploit this vulnerability by logging in to the device CLI with valid administrative (level 15) credentials and using crafted commands at the CLI prompt. A successful exploit could allow the attacker to execute arbitrary commands as root. |
| Improper Session Invalidation after Authentication Credential Change vulnerability in team-alembic ash_authentication allows an attacker holding a victim's session token to stay signed in after the victim resets or changes their password.
With apply_on_password_change? true, AshAuthentication.AddOn.LogOutEverywhere.Transformer adds OnPasswordChange to update actions as a global change guarded by where: changing(hashed_password). That condition is evaluated before the action's before_action hooks run, but AshAuthentication.Strategy.Password.HashPasswordChange sets the new hash in a before_action hook, so for every non-atomic password change the condition is false and the user's stored tokens are not revoked. The generated password reset action and the installer's change_password action are both non-atomic.
This issue affects ash_authentication: from 4.5.0 before 4.15.1 and from 5.0.0-rc.0 before 5.0.0-rc.16. |
| i18next-http-backend is a backend layer for i18next that loads translation resources in Node.js, browsers, and Deno. Prior to 4.0.4, interpolateUrl inserts unencoded language and namespace values into loadPath or addPath templates without rejecting ampersand (&), equals (=), or semicolon (;) query separators. When a template places lng or ns in the query string and an attacker controls that value, the attacker can add or override parameters on browser or server-side translation requests, potentially exposing other namespaces or tenants or changing backend behavior through the application's network position and credentials. This vulnerability is the result of an incomplete fix for CVE-2026-41691. This issue is fixed in version 4.0.4. |
