Security Advisories

hermes-agent has an Uncontrolled Resource Consumption issue

A security vulnerability has been detected in NousResearch hermes-agent up to 2026.4.30. This vulnerability affects the function _handle_webhook_request of the file gateway/platforms/feishu.py of the component Webhook Endpoint. Such manipulation leads to resource consumption. The attack can be launched remotely. The exploit has been disclosed publicly and may be used. The vendor was contacted early about this disclosure but did not respond in any way.

vLLM: OOM Denial of Service via Audio Decompression Bomb

### Summary vLLM's `/v1/audio/transcriptions` endpoint limits compressed upload size but not decoded PCM output. A 25MB OPUS file expands to ~14.9GB of float32 PCM at decode time. Tested on vLLM v0.19.0. ### Details `SpeechToTextProcessor` rejects uploads over `VLLM_MAX_AUDIO_CLIP_FILESIZE_MB` (default 25MB) based on compressed byte length, but the audio decoder in `audio.py` accumulates all decoded frames into memory with no size limit before returning: ```python # speech_to_text.py L184-189

vLLM: OOM Denial of Service via Audio Decompression Bomb

### Summary vLLM's `/v1/audio/transcriptions` endpoint limits compressed upload size but not decoded PCM output. A 25MB OPUS file expands to ~14.9GB of float32 PCM at decode time. Tested on vLLM v0.19.0. ### Details `SpeechToTextProcessor` rejects uploads over `VLLM_MAX_AUDIO_CLIP_FILESIZE_MB` (default 25MB) based on compressed byte length, but the audio decoder in `audio.py` accumulates all decoded frames into memory with no size limit before returning: ```python # speech_to_text.py L184-189

vLLM: OOM Denial of Service via Audio Decompression Bomb

### Summary vLLM's `/v1/audio/transcriptions` endpoint limits compressed upload size but not decoded PCM output. A 25MB OPUS file expands to ~14.9GB of float32 PCM at decode time. Tested on vLLM v0.19.0. ### Details `SpeechToTextProcessor` rejects uploads over `VLLM_MAX_AUDIO_CLIP_FILESIZE_MB` (default 25MB) based on compressed byte length, but the audio decoder in `audio.py` accumulates all decoded frames into memory with no size limit before returning: ```python # speech_to_text.py L184-189

vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant serving

## Summary Integer truncation of tensor dimensions in vLLM's GGUF dequantize kernels (`csrc/quantization/gguf/gguf_kernel.cu`) causes partial tensor processing. The output tensor is allocated at full size via `torch::empty` (uninitialized memory), but the dequantize CUDA kernel processes only a truncated number of elements. The unfilled portion of the output tensor retains whatever was previously in GPU memory. In multi-tenant inference deployments, this residual GPU memory may contain tensor d

vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant serving

## Summary Integer truncation of tensor dimensions in vLLM's GGUF dequantize kernels (`csrc/quantization/gguf/gguf_kernel.cu`) causes partial tensor processing. The output tensor is allocated at full size via `torch::empty` (uninitialized memory), but the dequantize CUDA kernel processes only a truncated number of elements. The unfilled portion of the output tensor retains whatever was previously in GPU memory. In multi-tenant inference deployments, this residual GPU memory may contain tensor d

vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant serving

## Summary Integer truncation of tensor dimensions in vLLM's GGUF dequantize kernels (`csrc/quantization/gguf/gguf_kernel.cu`) causes partial tensor processing. The output tensor is allocated at full size via `torch::empty` (uninitialized memory), but the dequantize CUDA kernel processes only a truncated number of elements. The unfilled portion of the output tensor retains whatever was previously in GPU memory. In multi-tenant inference deployments, this residual GPU memory may contain tensor d

vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations

## Summary Issue 1: EXIF orientation not normalized → The image orientation processed by the model differs from how humans view it, introducing interpretation bias. Issue 2: PNG tRNS not explicitly flattened before converting to RGB → After conversion, transparent/semi-transparent pixels are rendered unexpectedly, making otherwise subtle overlay elements visible and distorting the input content. (This attack is similar to AlphaDog: RGBA handling is already correct in vLLM, but since tRNS permi

vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations

## Summary Issue 1: EXIF orientation not normalized → The image orientation processed by the model differs from how humans view it, introducing interpretation bias. Issue 2: PNG tRNS not explicitly flattened before converting to RGB → After conversion, transparent/semi-transparent pixels are rendered unexpectedly, making otherwise subtle overlay elements visible and distorting the input content. (This attack is similar to AlphaDog: RGBA handling is already correct in vLLM, but since tRNS permi

vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations

## Summary Issue 1: EXIF orientation not normalized → The image orientation processed by the model differs from how humans view it, introducing interpretation bias. Issue 2: PNG tRNS not explicitly flattened before converting to RGB → After conversion, transparent/semi-transparent pixels are rendered unexpectedly, making otherwise subtle overlay elements visible and distorting the input content. (This attack is similar to AlphaDog: RGBA handling is already correct in vLLM, but since tRNS permi

vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels

## Summary All temperature validation gates use comparison operators (`<`, `>`), which silently evaluate to `False` for `NaN` and for positive `Infinity` in Python's IEEE 754 float semantics. Both values pass every guard and propagate to GPU sampling kernels, where they produce undefined behavior or CUDA errors that can crash the inference worker. Note: `-Infinity` is correctly caught. ## Root Cause `sampling_params.py:384`: ```python if 0 < self.temperature < _MAX_TEMP: # NaN → False; +Inf

vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels

## Summary All temperature validation gates use comparison operators (`<`, `>`), which silently evaluate to `False` for `NaN` and for positive `Infinity` in Python's IEEE 754 float semantics. Both values pass every guard and propagate to GPU sampling kernels, where they produce undefined behavior or CUDA errors that can crash the inference worker. Note: `-Infinity` is correctly caught. ## Root Cause `sampling_params.py:384`: ```python if 0 < self.temperature < _MAX_TEMP: # NaN → False; +Inf

vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels

## Summary All temperature validation gates use comparison operators (`<`, `>`), which silently evaluate to `False` for `NaN` and for positive `Infinity` in Python's IEEE 754 float semantics. Both values pass every guard and propagate to GPU sampling kernels, where they produce undefined behavior or CUDA errors that can crash the inference worker. Note: `-Infinity` is correctly caught. ## Root Cause `sampling_params.py:384`: ```python if 0 < self.temperature < _MAX_TEMP: # NaN → False; +Inf

Aider has an SSRF vulnerability through its AWS EC2 Metadata Endpoint

A security vulnerability has been detected in Aider-AI Aider 0.86.3.dev. This affects the function requests.get of the file api_docs.py of the component AWS EC2 Metadata Endpoint. The manipulation leads to server-side request forgery. The attack is possible to be carried out remotely. The exploit has been disclosed publicly and may be used. It is suggested to install a patch to address this issue. The pull request to fix this issue awaits acceptance.

Aider has an SSRF vulnerability through its AWS EC2 Metadata Endpoint

A security vulnerability has been detected in Aider-AI Aider 0.86.3.dev. This affects the function requests.get of the file api_docs.py of the component AWS EC2 Metadata Endpoint. The manipulation leads to server-side request forgery. The attack is possible to be carried out remotely. The exploit has been disclosed publicly and may be used. It is suggested to install a patch to address this issue. The pull request to fix this issue awaits acceptance.

Aider has an SSRF vulnerability through its AWS EC2 Metadata Endpoint

A security vulnerability has been detected in Aider-AI Aider 0.86.3.dev. This affects the function requests.get of the file api_docs.py of the component AWS EC2 Metadata Endpoint. The manipulation leads to server-side request forgery. The attack is possible to be carried out remotely. The exploit has been disclosed publicly and may be used. It is suggested to install a patch to address this issue. The pull request to fix this issue awaits acceptance.

Aider is vulnerable to Code Injection via editor_coder.run function

A security flaw has been discovered in Aider-AI Aider 0.86.3.dev. Affected by this vulnerability is the function editor_coder.run of the file auth.py of the component Architect Mode. Performing a manipulation results in code injection. Remote exploitation of the attack is possible. The exploit has been released to the public and may be used for attacks. The project was informed of the problem early through an issue report but has not responded yet.

Aider is vulnerable to Code Injection via editor_coder.run function

A security flaw has been discovered in Aider-AI Aider 0.86.3.dev. Affected by this vulnerability is the function editor_coder.run of the file auth.py of the component Architect Mode. Performing a manipulation results in code injection. Remote exploitation of the attack is possible. The exploit has been released to the public and may be used for attacks. The project was informed of the problem early through an issue report but has not responded yet.

Aider is vulnerable to Code Injection via editor_coder.run function

A security flaw has been discovered in Aider-AI Aider 0.86.3.dev. Affected by this vulnerability is the function editor_coder.run of the file auth.py of the component Architect Mode. Performing a manipulation results in code injection. Remote exploitation of the attack is possible. The exploit has been released to the public and may be used for attacks. The project was informed of the problem early through an issue report but has not responded yet.

vllm has Improper Resource Shutdown or Release

A vulnerability was identified in vllm-project vllm 0.19.0. This issue affects some unknown processing of the component OpenAI-compatible Serving Path. Such manipulation leads to denial of service. It is possible to launch the attack remotely. The exploit is publicly available and might be used. The pull request to fix this issue awaits acceptance.

vllm has Improper Resource Shutdown or Release

A vulnerability was identified in vllm-project vllm 0.19.0. This issue affects some unknown processing of the component OpenAI-compatible Serving Path. Such manipulation leads to denial of service. It is possible to launch the attack remotely. The exploit is publicly available and might be used. The pull request to fix this issue awaits acceptance.

vllm has Improper Resource Shutdown or Release

A vulnerability was identified in vllm-project vllm 0.19.0. This issue affects some unknown processing of the component OpenAI-compatible Serving Path. Such manipulation leads to denial of service. It is possible to launch the attack remotely. The exploit is publicly available and might be used. The pull request to fix this issue awaits acceptance.

wireshark-mcp vulnerable to arbitrary file write via export_objects when WIRESHARK_MCP_ALLOWED_DIRS is not configured

## Description ### Impact `wireshark-mcp` exposes a `wireshark_export_objects` MCP tool that accepts an attacker-controlled `dest_dir` parameter and passes it to tshark's `--export-objects` flag with **no mandatory path restriction**. The path sandbox (`_allowed_dirs`) is `None` by default and only activates when the environment variable `WIRESHARK_MCP_ALLOWED_DIRS` is explicitly set. In a default installation, any directory on the filesystem can be used as the export destination. **Affected

wireshark-mcp vulnerable to arbitrary file write via export_objects when WIRESHARK_MCP_ALLOWED_DIRS is not configured

## Description ### Impact `wireshark-mcp` exposes a `wireshark_export_objects` MCP tool that accepts an attacker-controlled `dest_dir` parameter and passes it to tshark's `--export-objects` flag with **no mandatory path restriction**. The path sandbox (`_allowed_dirs`) is `None` by default and only activates when the environment variable `WIRESHARK_MCP_ALLOWED_DIRS` is explicitly set. In a default installation, any directory on the filesystem can be used as the export destination. **Affected

Langflow vulnerable to injection

A vulnerability was detected in langflow-ai langflow up to 1.8.3. The impacted element is the function get_client_ip/install_mcp_config of the file src/backend/base/langflow/api/v1/mcp_projects.py of the component Model Context Protocol Configuration API. Performing a manipulation of the argument X-Forwarded-For results in injection. The attack may be initiated remotely. The exploit is now public and may be used. The vendor was contacted early about this disclosure but did not respond in any way

MS-Agent vulnerable to Command Injection

A Command Injection vulnerability in ModelScope's MS-Agent versions v1.6.0rc1 and earlier exists, allowing an attacker to execute arbitrary operating system commands through crafted prompt-derived input.

MCP Run Python has a Sandbox Escape & Server Takeover Vulnerability

### Impact **Critical Sandbox Escape & Server Takeover:** A critical security vulnerability exists in `mcp-run-python` due to a lack of isolation between the Python runtime (Pyodide) and the host JavaScript environment. The `runPython` and `runPythonAsync` functions execute Python code using Pyodide without restricting access to the JavaScript bridge. This allows any executed Python code—whether from a user or an AI model—to access the `js` module in Pyodide. Through this bridge, the Python cod

MCP Run Python Deno Sandbox Misconfiguration Allows SSRF Attacks via Localhost Access

### Impact **Server-Side Request Forgery (SSRF):** A security vulnerability exists in the `mcp-run-python` tool (specifically within the Pydantic-AI integration) due to an overly permissive Deno sandbox configuration. The tool configures the Deno runtime—which is intended to isolate the execution of untrusted Python code—with network permissions that include access to the host's loopback interface (`localhost`). Consequently, malicious Python code executed through the tool can bypass network is

mcp-kubernetes-server has a Command Injection vulnerability

`mcp-kubernetes-server` does not correctly enforce the `--disable-write` / `--disable-delete` protections when commands are chained. The server only inspects the first token to decide whether an operation is write/delete, which allows a read-like command to be followed by a write action using shell metacharacters (e.g., `kubectl version; kubectl delete pod <name>`). A remote attacker who can invoke the server may therefore bypass the intended write/delete restrictions and perform state-changing

vLLM Denial of Service via the best_of parameter

A vulnerability was found in the ilab model serve component, where improper handling of the best_of parameter in the vllm JSON web API can lead to a Denial of Service (DoS). The API used for LLM-based sentence or chat completion accepts a best_of parameter to return the best completion from several options. When this parameter is set to a large value, the API does not handle timeouts or resource exhaustion properly, allowing an attacker to cause a DoS by consuming excessive system resources. Thi

vLLM Denial of Service via the best_of parameter

A vulnerability was found in the ilab model serve component, where improper handling of the best_of parameter in the vllm JSON web API can lead to a Denial of Service (DoS). The API used for LLM-based sentence or chat completion accepts a best_of parameter to return the best completion from several options. When this parameter is set to a large value, the API does not handle timeouts or resource exhaustion properly, allowing an attacker to cause a DoS by consuming excessive system resources. Thi

vLLM Denial of Service via the best_of parameter

A vulnerability was found in the ilab model serve component, where improper handling of the best_of parameter in the vllm JSON web API can lead to a Denial of Service (DoS). The API used for LLM-based sentence or chat completion accepts a best_of parameter to return the best completion from several options. When this parameter is set to a large value, the API does not handle timeouts or resource exhaustion properly, allowing an attacker to cause a DoS by consuming excessive system resources. Thi

SQL injection in litellm

A blind SQL injection vulnerability exists in the berriai/litellm application, specifically within the '/team/update' process. The vulnerability arises due to the improper handling of the 'user_id' parameter in the raw SQL query used for deleting users. An attacker can exploit this vulnerability by injecting malicious SQL commands through the 'user_id' parameter, leading to potential unauthorized access to sensitive information such as API keys, user information, and tokens stored in the databas

SQL injection in litellm

A blind SQL injection vulnerability exists in the berriai/litellm application, specifically within the '/team/update' process. The vulnerability arises due to the improper handling of the 'user_id' parameter in the raw SQL query used for deleting users. An attacker can exploit this vulnerability by injecting malicious SQL commands through the 'user_id' parameter, leading to potential unauthorized access to sensitive information such as API keys, user information, and tokens stored in the databas

DesktopCommanderMCP is vulnerable to SSRF

A vulnerability was identified in wonderwhy-er DesktopCommanderMCP 0.2.37. This affects the function readFileFromUrl of the file src/tools/filesystem.ts of the component read_file. Such manipulation of the argument url leads to server-side request forgery. The attack may be performed from remote. The exploit is publicly available and might be used. The name of the patch is 53699bebba9950047bca16ac4dc8f0568f596aaa. It is best practice to apply a patch to resolve this issue.

DesktopCommanderMCP is vulnerable to SSRF

A vulnerability was identified in wonderwhy-er DesktopCommanderMCP 0.2.37. This affects the function readFileFromUrl of the file src/tools/filesystem.ts of the component read_file. Such manipulation of the argument url leads to server-side request forgery. The attack may be performed from remote. The exploit is publicly available and might be used. The name of the patch is 53699bebba9950047bca16ac4dc8f0568f596aaa. It is best practice to apply a patch to resolve this issue.

DesktopCommanderMCP is vulnerable to SSRF

A vulnerability was identified in wonderwhy-er DesktopCommanderMCP 0.2.37. This affects the function readFileFromUrl of the file src/tools/filesystem.ts of the component read_file. Such manipulation of the argument url leads to server-side request forgery. The attack may be performed from remote. The exploit is publicly available and might be used. The name of the patch is 53699bebba9950047bca16ac4dc8f0568f596aaa. It is best practice to apply a patch to resolve this issue.

DesktopCommanderMCP is vulnerable to SSRF

A vulnerability was identified in wonderwhy-er DesktopCommanderMCP 0.2.37. This affects the function readFileFromUrl of the file src/tools/filesystem.ts of the component read_file. Such manipulation of the argument url leads to server-side request forgery. The attack may be performed from remote. The exploit is publicly available and might be used. The name of the patch is 53699bebba9950047bca16ac4dc8f0568f596aaa. It is best practice to apply a patch to resolve this issue.

yii2-mcp-server has a Command Injection Issue

A flaw has been found in ArtMin96 yii2-mcp-server 1.0.2. This impacts the function yii_command_help/yii_execute_command of the file src/index.ts of the component MCP Interface. Executing a manipulation can lead to os command injection. The attack can be executed remotely. The exploit has been published and may be used. The project was informed of the problem early through an issue report but has not responded yet.

awwaiid mcp-server-taskwarrior vulnerable to command injection

A security vulnerability has been detected in awwaiid mcp-server-taskwarrior up to 1.0.1. This impacts the function server.setRequestHandler of the file index.ts. Such manipulation of the argument Identifier leads to command injection. The attack must be carried out locally. The exploit has been disclosed publicly and may be used. The name of the patch is 1ee3d282debfa0a99afeb41d22c4b2fd5a3148f2. Applying a patch is advised to resolve this issue. The vendor was contacted early, responded in a ve

fast-filesystem-mcp is vulnerable to command injection through handleGetDiskUsage function

A security flaw has been discovered in efforthye fast-filesystem-mcp up to 3.5.1. The affected element is the function handleGetDiskUsage of the file src/index.ts. Performing a manipulation results in command injection. The attack is possible to be carried out remotely. The exploit has been released to the public and may be used for attacks. The project was informed of the problem early through an issue report but has not responded yet.

OpenClaw: Telegram Webhook Missing Guess Rate Limiting Enables Brute-Force Guessing of Weak Webhook Secret

## Summary Telegram Webhook Missing Guess Rate Limiting Enables Brute-Force Guessing of Weak Webhook Secret ## Affected Packages / Versions - Package: `openclaw` - Affected versions: `<= 2026.3.24` - First patched version: `2026.3.25` - Latest published npm version at verification time: `2026.3.24` ## Details Telegram webhook auth previously rejected bad secrets but did not throttle repeated guesses, allowing brute-force attempts against weak webhook secrets. Commit `c2c136ae9517ddd0789d742

OpenClaw: Telegram Webhook Missing Guess Rate Limiting Enables Brute-Force Guessing of Weak Webhook Secret

## Summary Telegram Webhook Missing Guess Rate Limiting Enables Brute-Force Guessing of Weak Webhook Secret ## Affected Packages / Versions - Package: `openclaw` - Affected versions: `<= 2026.3.24` - First patched version: `2026.3.25` - Latest published npm version at verification time: `2026.3.24` ## Details Telegram webhook auth previously rejected bad secrets but did not throttle repeated guesses, allowing brute-force attempts against weak webhook secrets. Commit `c2c136ae9517ddd0789d742

OpenClaw: Telegram Webhook Missing Guess Rate Limiting Enables Brute-Force Guessing of Weak Webhook Secret

## Summary Telegram Webhook Missing Guess Rate Limiting Enables Brute-Force Guessing of Weak Webhook Secret ## Affected Packages / Versions - Package: `openclaw` - Affected versions: `<= 2026.3.24` - First patched version: `2026.3.25` - Latest published npm version at verification time: `2026.3.24` ## Details Telegram webhook auth previously rejected bad secrets but did not throttle repeated guesses, allowing brute-force attempts against weak webhook secrets. Commit `c2c136ae9517ddd0789d742