PYSEC-2026-3997

vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to crash the engine by submitting negative token IDs. A single request with a negative token ID triggers a CUDA device-side assertion that...

PYSEC-0000-CVE-2026-93592

vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to crash the engine by submitting negative token IDs. A single request with a negative token ID triggers a CUDA device-side assertion that...

virtualenv bash and fish activation scripts execute commands embedded in paths

The generated activate bash/zsh and activate.fish scripts interpolate ashlex.quote-ed value into a position that quotes it a second time. The extraquotes terminate the quoted run early and leave part of the value parsed asshell code, so a path containing shell metacharacters runs commands when a...

Downloaded seed wheels (pip/setuptools) are not integrity-checked before use

downloadwheel runs pip download and hands the result straight to theseeder, with nothing checking the bytes it gets back. The embedded pip andsetuptools wheels carry a BUNDLESHA256 that virtualenv checks on every load,but a wheel fetched over the network for the periodic-update feature or...

Command injection via --prompt in activate.bat (batch activator)

BatchActivator.quote returned its input unchanged, the only activator withno escaping at all. --prompt, the VIRTUALENVPROMPT environment variable,and the config file all set the prompt, and activate.bat writes it straightinto @set "VAR=value". A prompt containing a double quote closes that...

PYSEC-0000-CVE-2026-93436

vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. Remote attackers can submit requests with maxtokens=0 to exhaust decode-worker memory without bound until the worker restarts...

PYSEC-2026-3996

vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. Remote attackers can submit requests with maxtokens=0 to exhaust decode-worker memory without bound until the worker restarts...

virtualenv writes prompt values into pyvenv.cfg without sanitizing line boundaries, allowing configuration injection

pyvenv.cfg is a line-based format with no escape syntax. PyEnvCfg.writewrote values verbatim, while PyEnvCfg.readvalues parses the file withstr.splitlines. A value containing a line boundary therefore becameadditional configuration lines, and because reading is last-wins, the injectedkeys replace...

PYSEC-2026-3986

Apache Airflow Apache Kafka provider versions 1.15.0 before 2.0.0 resolve dotted-path strings found in a Kafka connection's extra field into Python callables via importstring, with no allowlist, and hand them to the confluent-kafka client which invokes them. Deployments that have enabled the Kafk...

PYSEC-2026-3985

vLLM before 0.28.0 contains a remote code execution vulnerability in the LlavaOnevision2 processor loader that ignores the trustremotecode parameter when loading remote processor classes. Attackers can craft a malicious model with arbitrary code in processingllavaonevision2.py that executes with...

PyTorch Lightning load_from_checkpoint has an insecure checkpoint deserialization

PyTorch-Lightning versions 2.6.0 and earlier contain an insecure deserialization vulnerability CWE-502 in the checkpoint loading mechanism. The LightningModule.loadfromcheckpoint method, which is commonly used to load saved model states, internally calls torch.load without setting the...

PYSEC-2026-3968

pytorch-lightning is vulnerable to Deserialization of Untrusted Data...

PyTorch Lightning denial of service vulnerability

A vulnerability in lightning-ai/pytorch-lightning version 2.3.2 allows an attacker to cause a denial of service by sending an unexpected POST request to the /api/v1/state endpoint of LightningApp. This issue occurs due to improper handling of unexpected state values, which results in the server...

Remote code execution in pytorch lightning

A remote code execution RCE vulnerability exists in the lightning-ai/pytorch-lightning library version 2.2.1 due to improper handling of deserialized user input and mismanagement of dunder attributes by the deepdiff library. The library uses deepdiff.Delta objects to modify application state base...

Compromise of PyTorch Lightning PyPi Package Versions

Security Advisory: Compromise of PyTorch Lightning PyPI Package Versions Published: 2026-04-30 Last Updated: 2026-05-12 Github Advisory: CVE-2026-44484We have identified a security incident affecting certain versions of one of our PyPI packages. What happenedWe have determined that one or more...

PYSEC-2026-3969

Code Injection in GitHub repository pytorchlightning/pytorch-lightning prior to 1.6.0...

pytorch-lightning vulnerable to Arbitrary File Write via /v1/runs API endpoint

A vulnerability in the /v1/runs API endpoint of lightning-ai/pytorch-lightning v2.2.4 allows attackers to exploit path traversal when extracting tar.gz files. When the LightningApp is running with the pluginserver, attackers can deploy malicious tar.gz plugins that embed arbitrary files with path...

PyTorch Lightning path traversal vulnerability

In lightning-ai/pytorch-lightning version 2.3.2, a vulnerability exists in the LightningApp when running on a Windows host. The vulnerability occurs at the /api/v1/uploadfile/ endpoint, allowing an attacker to write or overwrite arbitrary files by providing a crafted filename. This can lead to...

Open WebUI: Server-side fetches reach blocked and internal hosts via unvalidated HTTP redirect targets

SummaryOpen WebUI protects server-side web fetches with two controls: the operator's list of excluded hosts, and a check that refuses private and internal addresses. Neither control was applied to the destination of an HTTP redirect. On a deployment where redirect following is enabled, any...

SQLAdmin: Unvalidated sortBy parameter in `ModelView` bypasses `column_sortable_list`

SummaryModelView.sortquery uses the attacker-controlled sortBy list-view query parameter without checking it against the configured columnsortablelist allow-list. The value is resolved with getattrmodel, ... and fed into relationship joins and orderby, so a request can sort by any column of the...

Open WebUI: Any authenticated user can hang the server via message deletion in a cyclic chat tree

SummaryChat histories are stored as an unvalidated JSON object. After a message is deleted, the code that picks the chat's new current message walked down the childrenIds links without recording where it had already been. Any account with the default user role could store a chat whose messages li...

weasyprint Has Server-Side Request Forgery (SSRF)

Summaryurlfetcher is WeasyPrint's documented mechanism for restricting resource loading - applications use it to block file://, internal hosts, etc. when rendering untrusted input.Two writepdf channels ignore the document's urlfetcher and build a fresh default URLFetcher instead. A restrictive...

HTTPX2: Conflicting Content-Length and Transfer-Encoding headers can be auto-generated

SummaryHTTPX2 can automatically add a Content-Length header to a request that already contains a caller-supplied Transfer-Encoding header. The resulting HTTP/1.1 request contains both framing headers, which can create an ambiguous message boundary and enable request smuggling or connection...

Windows ML CLI: CORS misconfig enables localhost RCE

Case Description:MSRC Notes: Attachments: 1 files attached 1 mp4 Summary: The vulnerability lies in the 'serve/cliapi.py' component of the 'winml-cli' project, which exposes all winml CLI commands over HTTP without authentication. Although it binds to localhost by default, it sets 'alloworigins' ...

Open WebUI: Any authenticated user can hang the server via a cyclic chat message history

SummaryChat histories are stored as an unvalidated JSON object. The walk that reconstructs a chat's message chain detected repeats using each message's own id field while moving through the history by map key, so a message that simply omitted id was never recorded as visited. A history whose...

ESPHome Device Builder Dashboard: Unauthenticated dashboard access via the HA add-on ingress site bound to all interfaces

SummaryOn the Home Assistant add-on, the dashboard serves a trusted ingress site that skips authentication because the supervisor authenticates the request upstream. That site was binding 0.0.0.0. The add-on runs in host network mode for mDNS, so binding all interfaces also bound the host's LAN...

GitHacker: Path traversal in ref/hash parsing enables existence oracle and hex-fragment exfiltration via malicious .git server

SummaryGitHacker through 1.1.7 did not validate path segments parsed from attacker-controlled .git/HEAD before joining them onto its output directory. A malicious server could coerce GitHacker into reading arbitrary local files. Contents do not stream back wholesale, but the recovery loop turns a...

HTTPX2: Streaming response decompression does not bound peak memory (decompression amplification)

SummaryWhen decoding a compressed response body gzip, deflate, br, or zstd, HTTPX2 fully decompressed each network read before yielding content to the application. A small compressed input could therefore cause a large intermediate memory allocation, even when the application streamed the respons...

HTTPX2: Multipart part header injection via unvalidated file Content-Type and custom headers

SummaryHTTPX2 serializes the per-file Content-Type and custom headers supplied through the files= tuple API directly into the multipart/form-data body without validating custom header names or values. An attacker who can influence upload metadata passed to HTTPX2 can use CR or LF characters to...

Prowler: Stored XSS in HTML reports through unescaped cloud resource tags

SummaryProwler's HTML output formatter inserts finding.resourcetags into the generated report without HTML escaping. A cloud principal who can create or edit a resource tag in an account that is later scanned can store HTML or JavaScript in that tag. When another user opens the generated Prowler...

Prowler: Stored XSS in HTML reports through unescaped cloud resource tags

SummaryProwler's HTML output formatter inserts finding.resourcetags into the generated report without HTML escaping. A cloud principal who can create or edit a resource tag in an account that is later scanned can store HTML or JavaScript in that tag. When another user opens the generated Prowler...

vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections

Summaryvllm/transformersutils/processors/mimov2omni.py — the multimodal processor for MiMoV2OmniForCausalLM — issues requests.get... directly on user-supplied image and audio URL strings and Image.open... on user-supplied local paths, without the SSRF / allowedlocalmediapath checks that...

HTTPX2: Secure WebSocket traffic sent without TLS through SOCKS proxies

Summaryhttpcore2 does not start TLS for wss:// connections routed through a SOCKS5 proxy. The WebSocket opening handshake and all subsequent frames are sent in plaintext through the proxy path, despite the caller selecting the secure wss scheme.The transport flaw affects httpcore2 releases before...

HTTPX2: Secure WebSocket traffic sent without TLS through SOCKS proxies

Summaryhttpcore2 does not start TLS for wss:// connections routed through a SOCKS5 proxy. The WebSocket opening handshake and all subsequent frames are sent in plaintext through the proxy path, despite the caller selecting the secure wss scheme.The transport flaw affects httpcore2 releases before...

vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m

SummaryThe fix for GHSA-rwxx-mrjm-wc2m "ReDoS via structuredoutputs.regex compiled without timeout" wrapped the regex compile in the xgrammar and outlines backends with compileregexwithtimeout and, for outlines, validateregexisbuildable. The lm-format-enforcer backend was left unguarded: it...

vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds

SummaryThe /v1/completions/derender and /v1/chat/completions/derender endpoints accept caller-supplied GenerateResponse objects and postprocess every nested choices.tokenids list directly. Unlike the normal render/generate path, derender does not enforce model context length, resolved maxtokens,...

vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error Messages

SummaryWhen the vLLM API receives a malformed request e.g., invalid JSON or missing required fields, FastAPI raises a Pydantic RequestValidationError. The validationexceptionhandler in vllm/entrypoints/openai/serverutils.py converts this exception to a string via strexc, which includes the intern...

HTTPX2: Quadratic SSE line buffering can cause CPU denial of service

SummaryHTTPX2's Server-Sent Events SSE parser repeatedly copied and rescanned buffered text when a server split one unterminated line across many response chunks. The total work grows quadratically with the length of the line. An attacker-controlled or compromised SSE endpoint can exploit this...

GitPython: TagReference.create positional reference bypasses kwargs-only --file guard, enabling arbitrary file read (incomplete fix of 3af0c251)

SummaryTagReference.create forwards a caller-influenced positional reference value into git tag without it ever being inspected by the unsafe-option guard, allowing an arbitrary file read the file's contents are returned in-band as the annotated tag message. This is an incomplete-fix bypass of...

vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt parts

Executive SummaryThe follow-up protection for CVE-2025-62164 is incomplete at vLLM revision 26587f9519e22a5c4549ead7595ad9ca3229c4fd. It wraps serialized prompt-embedding reconstruction and dense conversion in torch.sparse.checksparsetensorinvariants, but PyTorch 2.11.0 implements that context wi...

vLLM: Cross-User Data Leak Vulnerability

SummaryAn integer overflow in the actandmulkernel kernel can cause the output of one user request to be incorporated into the response of another request within the same inference batch. Under certain conditions, the last request in a batch can receive a partial or complete copy of the first user...

Omnigent: Unvalidated os_env.cwd in agent bundle yields arbitrary host filesystem access on runners without OMNIGENT_RUNNER_WORKSPACE

SummaryAn authenticated, non-admin user can obtain arbitrary host-filesystem read/write and host environment-secret disclosure on an Omnigent runner by uploading an agent bundle whose osenv.cwd points outside any intended workspace e.g. / or /home/. The cwd field is taken verbatim from the bundle...

Banks: Path traversal in `DirectoryPromptRegistry.set()` allows arbitrary file write outside the registry root

SummaryDirectoryPromptRegistry.set interpolates the attacker-controllable Prompt.name into a Path expression with no canonicalization. An application that derives the prompt name from request data lets a caller write attacker-controlled bytes outside the configured registry directory...

Material for MkDocs: DOM XSS in search suggestions via query parameter

ImpactMaterial for MkDocs 7.2.0 through 9.7.6 contains a DOM-based cross-site scripting vulnerability in the optional search.suggest feature. A crafted q URL parameter could execute JavaScript in the documentation site's origin after user interaction. PatchesThe issue is fixed in Material for...

unstructured: Server-Side Request Forgery in the URL-based partitioning

SummaryServer-Side Request Forgery in unstructured. The url= argument of partition, partitionhtml, and partitionmd is fetched via requests.get with no host validation. The response body is returned as Element text, so this is a full-read SSRF — attackers reach loopback admin APIs, internal HTTP...

Omnigent: Shared Agent Bundle Overwrite Leads to Authenticated Runner RCE

SummaryAn authenticated user with edit access to their own session can overwrite a shared/template agent by uploading a full agent bundle through PUT /sessions/sessionid/agent.Shared/template agents are shown as not MCP-editable, but this upload path still accepts a replacement bundle. By adding ...

sqlparse: Reindentation of tuple lists causes near-cap quadratic CPU consumption

SummaryWhen SQL is formatted with reindentation enabled, ReindentFilter repeatedly rebuilds prefixes of the current statement to calculate token offsets. An attacker who controls SQL sent to this opt-in formatting path can supply a parenthesized tuple list that remains just below the grouping-tok...

Omnigent Guardrail policy bypass: shell-command parser fails open in policies/builtins/_shell.py

Reporter: Aaron / Aeon — autonomous security agent https://github.com/aaronjmars/aeonProject: omnigent-ai/omnigent v0.1.0 Databricks — meta-harness running Claude Code / Codex / Pi "in check with policies and sandboxing"Component: omnigent/policies/builtins/shell.py shared parser → consumed by...

Omnigent: Uploaded Agent Bundle Allows Authenticated Runner RCE via Python Callable Tools

SummaryAn authenticated user can upload a crafted agent bundle that defines a server-side Python callable tool. The server validates the uploaded bundle, but it does not block dangerous callable: paths in untrusted user-provided agent configs.When the tool is invoked, the runner imports and...

pypdf: Possible infinite loop for TreeObject.insert_child

ImpactAn attacker who uses this vulnerability can craft a PDF which leads to an infinite loop. This requires a usually writing code path where TreeObject.insertchild is involved. PatchesThis has been fixed in pypdf==6.16.0. WorkaroundsIf you cannot upgrade yet, consider applying the changes from ...