Open WebUI: Cross-user file disclosure via /api/chat/completions image_url field

summaryPOST /api/chat/completions accepts an imageurl.url value that, when it does NOT start with http://, https://, or data:image/, is interpreted as a file id and resolved against the global file table with no ownership check. An authenticated user can therefore set imageurl.url to another user...

Open WebUI: Cross-origin postMessage confirmation bypass via action:submit

SummaryThe chat message listener allows non-same-origin input:prompt and action:submit messages, so an external site can set prompt text and trigger submitPrompt in an authenticated victim session. I validated this with a cross-origin attacker page that auto-posted messages and caused unauthorize...

Open WebUI: Stored XSS to Account Takeover via Model Profile Images

Stored XSS to Account Takeover via Model Profile Images in Open WebUIAffected: Open WebUI = 0.9.5Bypass of: GHSA-3wgj-c2hg-vm6q, GHSA-3856-3vxq-m6fc--- TL;DROpen WebUI patched SVG XSS in user profile images and webhook profile images but forgot to apply the same fix to model profile images. The...

yt-dlp: File Downloader cookie leak with curl

SummaryIf curl is used an external downloader for yt-dlp, cookies may be leaked to an unintended host upon HTTP redirect or when the host for download fragments differs from their parent manifest's.This is the equivalent to GHSA-v8mc-9377-rwjj for the curl downloader. The vulnerable behavior is...

vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router

vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via the Anthropic API routerResearcher: Kai Aizen — SnailSploit @SnailSploit, Adversarial & Offensive Security ResearchSeverity: CVSS 3.1 5.3 Medium AV:N/AC:L/PR:N/UI:N/S:U/C:L/I:N/A:NTarget: https://github.com/vllm-project/vllm---...

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

SummaryIssue 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...

OpenStack Nova: Nova scheduler hint injection bypasses Placement resource claims and scheduling constraints

Affects- Nova: =18.0.0 =32.0.0 =33.0.0 33.0.2 DescriptionErichen from the Institute of Computing Technology, Chinese Academy of Sciences reported that Nova's server create API does not strip internal scheduler hints. An authenticated user can bypass Placement resource claims and scheduling...

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

SummaryAll 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...

yt-dlp: Arbitrary code execution via manifest downloads with aria2c

SummaryIf aria2c is used as an external downloader for a fragmented manifest format such as an HLS/DASH stream, yt-dlp passes insufficiently sanitized input to aria2c that allows an attacker to perform an arbitrary file write. On Windows platforms, this can lead to immediate arbitrary code...

yt-dlp: Dangerous file type creation via insufficient filename sanitization (Bypass of CVE-2024-38519)

SummaryA vulnerability exists in yt-dlp that allows a remote attacker to write arbitrary OS-shortcut files such as .desktop, .url, .webloc to the user's filesystem, bypassing the remediation for CVE-2024-38519. DetailsThe fix for CVE-2024-38519 enforced an allowlist for file extensions, in order ...

vLLM: OOM Denial of Service via Audio Decompression Bomb

SummaryvLLM'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. DetailsSpeechToTextProcessor rejects uploads over VLLMMAXAUDIOCLIPFILESIZEMB default 25MB based on...

Open WebUI IDOR: Calendar event re-parenting allows writing events into another user's calendar

SummaryPOST /api/v1/calendars/events/eventid/update validates that the caller has write access to the calendar the event currently belongs to, but does not validate the destination calendarid supplied in the request body. The model layer then persists the new calendarid unconditionally.A regular...

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

SummaryInteger truncation of tensor dimensions in vLLM's GGUF dequantize kernels csrc/quantization/gguf/ggufkernel.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 numb...

Langflow: Path Traversal in Knowledge Bases API via Creation Endpoint

SummaryLangflow is vulnerable to Path Traversal in the Knowledge Bases API POST /api/v1/knowledgebases. This occurs because user-supplied knowledge base names are used directly to create file paths without proper sanitization or containment checks. An authenticated attacker can exploit this flaw ...

Tornado has out-of-bounds memory access via C extension

SummaryTornado's optional native extension tornado.speedups implements websocketmask without validating that the mask argument is exactly four bytes long. The C function reads four bytes from mask unconditionally, even when Python passes a shorter byte string. This can read beyond the provided...

python-multipart: Semicolon treated as querystring field separator enables parameter smuggling

SummaryQuerystringParser treated ; as a field separator in application/x-www-form-urlencoded bodies, in addition to &. The WHATWG URL standard, modern browsers, and Python's urllib.parse since the CVE-2021-23336 fix treat only & as a separator. This creates a parser differential: the same bytes a...

Tornado: Authorization header forwarded across cross-origin redirects in SimpleAsyncHTTPClient

SummaryWhen SimpleAsyncHTTPClient follows a 3xx redirect, it shallow-copies the original HTTPRequest, rewrites the URL, decrements maxredirects, and removes only the Host header. It does not clear Authorization, authusername, authpassword, or authmode when the redirect target changes origin.As a...

tornado AsyncHTTPClient accumulates decompressed chunks without size limit (gzip bomb)

Tornado's gzip decompression routines work in limited-size chunks, but have no overall limit for the total size of decompressed chunks that they will accumulate There has always been a limit for the total compressed size. This allows a malicious server to consume effectively unlimited amounts of...

Galaxy NG: command injection vulnerability

A command injection vulnerability was found in galaxyng. The dogitcheckout function in the legacy role import API v1 interpolates unsanitized git ref names branch/tag names into shell commands executed via subprocess.run with shell=True. An authenticated user who controls a git repository can...

LangChain: Path traversal and sandbox escape in LangChain file-search middleware and loaders

SummarySeveral LangChain components that resolve filesystem paths or expand search patterns do not consistently confine the resolved path to the intended root directory. Affected behaviors include: a file-search agent middleware that validates a starting directory but not the search pattern or th...

python-multipart: Quadratic-time querystring parsing with semicolon separators causes CPU denial of service

SummaryWhen parsing application/x-www-form-urlencoded bodies, QuerystringParser located the field separator with a two step lookup: it first scanned the entire remaining buffer for &, and only when no & existed anywhere ahead did it fall back to scanning for ;. For a body that uses ; as the...

pypdf: Inefficient decoding of FlateDecode PNG predictor streams

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to long runtimes. This requires accessing a stream which uses the /FlateDecode filter with a PNG predictor. PatchesThis has been fixed in pypdf==6.12.2. WorkaroundsIf you cannot upgrade yet, consider applying the changes fr...

pypdf: Possible infinite loop when retrieving fonts for layout-mode text extraction

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to an infinite loop. This requires extracting the text in layout mode. PatchesThis has been fixed in pypdf==6.13.0. WorkaroundsIf you cannot upgrade yet, consider applying the changes from PR 3830...

python-multipart: Content-Disposition parameter smuggling via RFC 2231/5987 extended parameters

Summaryparseoptionsheader parsed Content-Disposition and Content-Type headers with email.message.Message, which transparently applies RFC 2231/5987 decoding. The extended parameter syntax filename=charset'lang'value, name=..., and the filename0/filename1 continuation form is decoded and surfaced...

pypdf: Possible infinite loop when processing outlines/bookmarks in writer

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to an infinite loop. This requires merging a file with outlines into a writer. PatchesThis has been fixed in pypdf==6.13.0. WorkaroundsIf you cannot upgrade yet, consider applying the changes from PR 3830...

pypdf: Manipulated XMP metadata streams can exhaust RAM

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to large memory usage. This requires parsing large XMP metadata, possibly with lots of unnecessary elements. PatchesThis has been fixed in pypdf==6.12.1. WorkaroundsIf you cannot upgrade yet, consider applying the changes...

python-multipart: Negative Content-Length in parse_form buffers the entire body in memory

Summaryparseform did not validate the Content-Length header before using it to bound its chunked read of the request body. A negative Content-Length turned the bounded read into a read-until-EOF, so the entire body was loaded into memory in a single read instead of in fixed-size chunks...

pypdf: Possible large memory usage for form XObjects during text extraction

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to large memory usage. This requires extracting the text of a page which contains a form XObject with self-references. PatchesThis has been fixed in pypdf==6.12.2. WorkaroundsIf you cannot upgrade yet, consider applying the...

pypdf: Possible long runtimes for zero-only width values in cross-reference streamsuntimes for zero-only width values in cross-reference streams

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to long runtimes. This requires cross-reference streams with /W 0 0 0 values and large /Size values. PatchesThis has been fixed in pypdf==6.12.0. WorkaroundsIf developers are unable to upgrade their apps immediately, they...

PDM: Project-Local State and Config Writes Follow Symlinks

SummaryPDM writes several project-local state or configuration files without symlink protection. If a malicious repository places those files as symlinks, local PDM operations can overwrite the symlink targets.This creates an arbitrary file clobber primitive relative to the privileges of the...

python-zeroconf: Unbounded TC-deferred queue allows LAN-local memory exhaustion via spoofed-source flood

ImpactAsyncListener.handlequeryordefer retained every truncated TC-bit incoming query in self.deferredaddr and armed a per-addr timer in self.timersaddr that flushed the reassembled query within 500 ms RFC 6762 §18.5. Neither the per-addr list nor the number of distinct addr keys was capped, and...

PDM: Project-Controlled `.pdm-plugins` Content Executes Before CLI Parsing

SummaryPDM automatically loads project-local plugin paths from .pdm-plugins during Core initialization. Because this path is added via site.addsitedir, attacker-controlled .pth files inside the project plugin directory are processed and can execute Python code before normal CLI handling begins.Th...

pypdf: Possible large memory usage for large offsets for layout mode text

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to large memory usage. This requires extracting text in layout mode with large character offsets. PatchesThis has been fixed in pypdf==6.12.0. WorkaroundsIf developers are unable to immediately upgrade, they should consider...

WsgiDAV encoded dot segments can escape filesystem share roots

ImpactWsgiDAV 4.3.3 can allow a WebDAV request path containing an encoded parent-directory segment to escape the configured filesystem share root in a specific path layout. PatchesThe issue is fixed with version 4.3.4. PreconditionsThe practical impact depends on the deployment.The deployment use...

Kolibri has Unauthenticated Server-Side Request Forgery (SSRF) in RemoteFacilityUserViewset

SummarySeveral Kolibri API endpoints accept an unvalidated baseurl parameter and fetch attacker-controlled URLs from the Kolibri server, reflecting the response body back to the caller. The original report identified two endpoints on the RemoteFacilityUser viewsets; remediation review found two...

PDM wheel installation leads to Path Traversal via overridden write_to_fs

InstallDestination.writetofs in src/pdm/installers/installers.py overrides the base class to add symlink/hardlink support but replaces the safe pathwithdestdir which validates via Path.resolve + isrelativeto with a bare os.path.join that performs no path validation. A malicious wheel with travers...

Bugsink: Project scoping missing in sourcemap and debug-file lookup

SummaryBugsink before 2.2.0 resolved sourcemaps and debug files by debug ID without scoping that lookup to the project that owned the uploaded metadata. An authenticated user with access to one project could cause event processing in that project to use sourcemap/debug-file metadata uploaded for...

Bugsink: Issue bulk actions can affect another project’s issue if its UUID is known

DescriptionBugsink’s issue list supports bulk actions such as resolving or muting selected issues. In affected versions, the issue list view authorizes access through the project in the URL, but applies the requested bulk action to the submitted issue IDs without also requiring those issues to...

Litestar has HTML Injection Through its CSRF Token

OverviewLitestar instances which use a template engine in conjunction with CSRF protection are vulnerable to HTML Injection which can be escalated to Cross Site Scripting due to the contents of the CSRF cookie being excluded from automatic escaping by the template engine when configured inline wi...

Vantage6: Set admin user and password from environment or configuration

ImpactVantage6 currently provides an initial user with username root and password root. This is not ideal for the following reasons:- Attackers know that almost all vantage6 servers have a user with username root that probably has admin rights- The initial password is very weak and it is possible...

vantage6 node has an Improper Access Control issue

ImpactMalicious algorithms can potentially access other algorithms input and output files. PatchesTodo WorkaroundsVerify and restrict the algorithm containers that are allowed to run on your node. See here on how to do this. Referenceshttps://docs.vantage6.ai/usage/running-the-node/security For...

Litestar: AllowedHostsMiddleware bypasses host validation via client-controlled X-Forwarded-Host header

SummaryAllowedHostsMiddleware trusts the X-Forwarded-Host header as a fallback when the Host header is absent. Since X-Forwarded-Host is a client-controllable header, an attacker can bypass the allowed hosts validation by omitting the Host header and supplying an X-Forwarded-Host header set to a...

Dulwich doesn't sanitize commit subjects in `porcelain.format_patch`

Impactdulwich.porcelain.formatpatchoutdir=... derives each patch filename from the commit's subject line. Prior to this fix, getsummary only replaced spaces with dashes - path separators /, \, parent-directory components .., and other filename-hostile characters e.g. : were preserved verbatim and...

Bugsink: DOS using large numbers of event tags

SummaryIn affected versions, Bugsink stores every tag supplied with an incoming event. An event with an unusually large numberof custom i.e. supplied by an attacker tags can therefore make ingestion spend more time than intended writing tag rows.Bugsink uses a single-writer database architecture...

Dulwich has unbounded memory allocation in receive-pack from crafted thin packs

ImpactAn uncontrolled-resource-consumption memory exhaustion denial-of-service vulnerability CWE-400 / CWE-789.A client with push access could push a tiny crafted thin pack 174 bytes whose delta header declares a huge destsize. When dulwich ingested it via addthinpack / applydelta, it would...

Bugsink: Issue event views can show an event from another project if its UUID is known

DescriptionBugsink issue event pages accept a direct event identifier from the URL and, in affected versions, look up that event without also requiring it to belong to the issue in the URL.This is a project-boundary authorization issue: a logged-in user with access to one project can view another...

Docling Core: Insufficient validation of image reference URIs

ImpactIn versions = 2.5.0, = 2.74.1 WorkaroundsIf upgrading is not immediately possible:- reject file: and data: image references from untrusted input- allow only approved local or remote image sources- apply input size and memory limits to processing workers References- Fix release: v2.74.1...

malla: Stored XSS via Meshtastic node names in multiple frontend pages

Node names longname, shortname received via MQTT are stored in SQLite without sanitization and rendered into the DOM without escaping.Any participant on a public Meshtastic MQTT broker can set a malicious node name that executes JavaScript in the browser of every Malla dashboard visitor.Affected...

Docling Core: Unsafe remote filename resolution

ImpactIn versions = 1.5.0, = 2.74.1 WorkaroundsIf upgrading is not immediately possible, avoid passing untrusted URLs into remote fetch functionality. References- Fix release: v2.74.1...

OpenMed vulnerable to remote code injection through privacy-filter model loading path

OpenMed before 1.5.2 contains a remote code execution vulnerability in the PII privacy-filter model loading path. The privacy-filter dispatcher used broad substring matching on the user-supplied modelname parameter, allowing a value such as attacker/foo-privacy-filter-bar to route through a path...