Odysseus before commit bf325f6 contains a missing authorization vulnerability that allows authenticated non-admin users to manage server-wide embedding backend configuration by invoking endpoint management routes that verify session authentication but omit the admin authorization guard. Attackers can supply an attacker-controlled URL to overwrite the embedding backend persisted in the endpoint configuration file and process environment, causing all subsequent embedding operations including chat messages, RAG queries, memory entries, and vault text to be transmitted in plaintext to the attacker-controlled destination, or delete the endpoint configuration to deny embedding service to all users.

OpenSIPS is a Session Initiation Protocol (SIP) server implementation. In versions prior to 3.6.6 and 4.0.0-rc1, the TCP message framing layer parses the Content-Length header using unsigned int arithmetic with no overflow check. When an attacker sends a Content-Length value that overflows unsigned int (e.g., 4294967296), the framing layer computes a wrapped-around value (e.g., 0) and splits the TCP stream at the wrong boundary, causing the body of the first SIP message to be processed as a separate message and enabling SIP message smuggling. Because Content-Length is parsed in the transport layer before authentication, an unauthenticated, network-based attacker can smuggle arbitrary SIP messages over any TCP-based transport (proto_tcp, proto_tls, proto_ws, proto_wss) on any instance with TCP enabled, with no routing-script preconditions. This allows smuggled messages to bypass front-end SBC/proxy security policies, inherit the connection's authentication context, and evade rate limiting. This issue has been fixed in versions 3.6.6 and 4.0.0-rc1.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. From 0.9.6 until 0.11.0, the built-in knowledge search path in backend/open_webui/tools/knowledge_fs.py and backend/open_webui/tools/builtin.py let a chat participant choose a pattern used to grep knowledge files. Patterns containing regex metacharacters were compiled with Python's backtracking re engine and run against every line of every reachable file with no time limit, so a crafted pattern such as (x|x)*y and one matching uploaded file line can pin one CPU core and block the event loop. This causes availability impact for every other user of the affected worker. This issue is fixed in 0.11.0.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. In 0.10.2 and earlier, the GET /api/v1/tools/, GET /api/v1/tools/list, and GET /api/v1/tools/id/{id} endpoints in backend/open_webui/routers/tools.py returned full Python tool source to authenticated non-admin read-only users. ToolResponse deliberately omitted source and specs, but ToolUserResponse permitted extra fields and handlers spread a full tool model dump into the response, re-admitting omitted fields. A non-admin with a read grant can obtain another user's server-side tool source, which commonly embeds hard-coded API keys, credentials, and internal service URLs. This issue is fixed in 0.11.0.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. Prior to 0.11.0, Open WebUI resolved a hostname during URL validation and rejected private, loopback, and link-local addresses, but the HTTP clients resolved the hostname again at connection time. An authenticated attacker who controlled authoritative DNS for a submitted hostname could answer with a public address during validation and an internal one during connection, reaching cloud metadata, loopback admin APIs, or internal services through URL ingest, chat image_url fetches, image editing, or OAuth profile-picture fetches, with most paths returning the response to the attacker and the OAuth path forwarding the OAuth access token. This issue is fixed in 0.11.0.

OpenSIPS is a Session Initiation Protocol (SIP) server implementation. In versions 4.0.0 and prior, processing a SIP message with a header name longer than 255 bytes causes a stack buffer overflow when sip_to_json() is called in the routing script. Function sip_to_json() (modules/sipmsgops/sipmsgops.c) copies SIP header names into a fixed 255-byte stack buffer without bounds checking, performing a memcpy of the full header-name length even though the SIP parser imposes no such limit (a header name can be roughly 65000 bytes). As a result, when a routing script calls sip_to_json(), a SIP message with a header name longer than 255 bytes triggers a stack buffer overflow in which both the length and content of the overwrite are attacker-controlled, corrupting the saved frame pointer and return address. A single unauthenticated UDP packet to the SIP port (5060) can crash the process or, on builds without stack protections, hijack the return address to achieve remote code execution. This affects deployments whose routing script invokes sip_to_json(). This issue was not fixed at the time of publication.

MaxSite CMS contains a remote code execution vulnerability that allows unauthenticated attackers to inject arbitrary PHP code into the application configuration file by submitting crafted POST requests to the install endpoint after installation is complete. Attackers can supply a malicious db_dbprefix value containing a single quote to break out of a PHP string literal in application/config/database.php, appending attacker-controlled PHP statements that are executed by the web server on every subsequent request, resulting in persistent unauthenticated remote code execution as the web-server process user.

Flowise is a drag-and-drop user interface for building customized large language model (LLM) flows. Prior to 3.1.3, Flowise GET /api/v1/upsert-history returns the entire server-wide upsert history instead of being scoped to the requesting user, tenant, or workspace. The response can exceed 100MB and includes sensitive configuration data, including Vector Store settings such as Qdrant Server URL and collection name. The observed behavior indicates missing or insufficient authorization checks, workspace/project/tenant isolation, and pagination or limits, exposing integration parameters and infrastructure details that may enable further targeted attacks. This issue is fixed in version 3.1.3.

pdm is a Python package and dependency manager supporting the latest PEP standards. In versions prior to 2.27.0, pdm 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 invoking user. Config.__init__() resolves the project-local pdm.toml path and _save_config() writes to the resolved target. If PROJECT_ROOT/pdm.toml is a symlink to another file, pdm config -l ... updates the target file instead of refusing the write. The same general problem exists for other project-local persistence paths that are written directly with no lstat / O_NOFOLLOW protection. For the pdm.toml PoC specifically, the target file must already contain parseable TOML. Otherwise the load step fails before the write path is reached. That parser constraint does not apply to the .pdm-python or .python-version sinks. This issue has been fixed in version 2.27.0.

NVIDIA Dynamo for Linux contains a vulnerability where an attacker may cause improper limitation of a pathname to a restricted directory by supplying a crafted local path in a multimodal request. A successful exploit of this vulnerability might lead to information disclosure.

NVIDIA Dynamo for Linux contains a vulnerability in the image loading component where an attacker may cause improper limitation of a pathname to a restricted directory. A successful exploit of this vulnerability might lead to information disclosure.

NVIDIA Dynamo for Linux contains a vulnerability in the multimodal embedding cache, where an attacker could cause a hash collision by submitting images that share an identical pixel byte sequence but have different dimensions. A successful exploit of this vulnerability might lead to data tampering.

Flowise is a drag & drop user interface to build a customized large language model flow. Prior to 3.1.3, the mitigation for CVE-2025-8943 blocked -y and --yes flags on npx, but packages/components/nodes/tools/MCP/core.ts denied only PATH, LD_LIBRARY_PATH, DYLD_LIBRARY_PATH, and NODE_OPTIONS by exact environment-variable name. Because npm reads configuration from npm_config_* variables, setting npm_config_yes=true reproduced --yes behavior without using a blocked flag, causing npx to auto-install and execute the named package when a Custom MCP server launched. This issue is fixed in version 3.1.3.

An issue was discovered in Django 5.2 before 5.2.17 and 6.0 before 6.0.8. `django.utils.translation.check_for_language()` is subject to a potential denial-of-service attack when given many distinct, very long language codes, which are retained as keys in an in-memory cache and consume process memory. Such codes reach the function through the `django.views.i18n.set_language()` view, which is not routed by default. The consumed memory is bounded, since request data is limited by the `DATA_UPLOAD_MAX_MEMORY_SIZE` setting (default 2.5 MB) and the cache holds a fixed maximum number of entries. Earlier, unsupported Django series (such as 5.1.x, 5.0.x, and 4.2.x) were not evaluated and may also be affected. Django would like to thank Jaeyoung Jang for reporting this issue.

An issue was discovered in Django 5.2 before 5.2.17 and 6.0 before 6.0.8. GeoDjango spatial lookups optimistically parse the right-hand-side value as a raster by passing it to the `django.contrib.gis.gdal.GDALRaster` constructor. Any value used in a spatial lookup against a `GeometryField` or `RasterField` reaches this constructor, including untrusted input, for example a spatial-field filter submitted through the Django admin changelist query string by a staff user with view permission. A `dict`, or a `str` holding its JSON representation, is opened in write mode regardless of the constructor's `write=False` default, allowing a file with an attacker-chosen name and contents to be written through a file-backed GDAL driver. Any other `str` is treated as a datasource, allowing an outbound network request through a GDAL virtual filesystem handler. Writing a file to a location later imported by the application can result in remote code execution. Earlier, unsupported Django series (such as 5.1.x, 5.0.x, and 4.2.x) were not evaluated and may also be affected. Django would like to thank Bence Nagy, localhost-detect, and kimchunbok_ for reporting this issue.

LAMP Rapid Development Platform through 5.6.2, fixed in commit 84b0c27, contains a remote code execution vulnerability in GlueFactory that executes unsandboxed Groovy scripts from database template fields without compilation restrictions or whitelisting. Attackers can write or influence the script field via message template endpoints to execute arbitrary Groovy code and OS commands on the backend server.

The fix released in jackson-core 2.18.6 and 2.21.1 for CVE-2026-18401 (GHSA-72hv-8253-57qq, number length constraint bypass in the non-blocking parser) is incomplete. This record covers the remaining bypass. The earlier fix wired validateIntegerLength() into a new _setIntLength() helper and invoked it wherever the integer portion of a number is decided: a terminator byte arrives, a '.' or 'e'/'E' is seen, or input ends inside a fully buffered value. It was not invoked on the attacker-relevant path where the parser runs out of input while still inside the MINOR_NUMBER_INTEGER_DIGITS minor state and returns NOT_AVAILABLE to the caller. As a result, an attacker who streams JSON to a non-blocking parser in many small chunks, without ever sending a terminator byte, keeps the parser inside MINOR_NUMBER_INTEGER_DIGITS indefinitely. _textBuffer.expandCurrentSegment() grows the accumulator on every chunk while validateIntegerLength() is never called. The accumulator is bounded only by maxStringLength (20 MiB by default) rather than by maxNumberLength (1000 by default), an amplification of roughly 20,000x over the documented limit. Because Java char values occupy two bytes, a single connection can be driven to approximately 40 MiB of heap before the validator finally fires when the value completes. The equivalent fraction-path code is correct: _finishFloatFraction() calls _setFractLength() before its NOT_AVAILABLE return. The missing call affects the integer-digit paths in _startPositiveNumber(), _startNegativeNumber() and _finishNumberIntegralPart() in NonBlockingUtf8JsonParserBase. Impact: reactive frameworks such as Spring WebFlux/Reactor, Quarkus, Helidon and Vert.x feed inbound HTTP or gRPC bytes to the async parser as they arrive, which is precisely the chunked-feed shape required. Operators who set StreamReadConstraints.maxNumberLength expecting it to cap memory per number value do not get that guarantee; memory accumulates per concurrent connection and attacker-controlled concurrency can exhaust the JVM heap. The synchronous parsers (UTF8StreamJsonParser, ReaderBasedJsonParser) and the async parser operating on complete input are not affected. Exploitation requires only the ability to stream data to a parsing endpoint; no privileges or user interaction are needed. This issue affects com.fasterxml.jackson.core:jackson-core from version 2.15.0 through 2.18.7, and from 2.19.0 through 2.21.3, and tools.jackson.core:jackson-core from 3.0.0 through 3.1.3. Versions prior to 2.15.0 are not affected, because StreamReadConstraints -- which defines the maxNumberLength setting -- was first introduced in jackson-core 2.15.0, so no such constraint exists to be bypassed in earlier releases. Note that GHSA-r7wm-3cxj-wff9 states the affected 2.x range without a lower bound. The 2.22.x and 3.2.x release lines are not affected: those branches were created after the fix commit landed on 2026-05-21 and therefore contain it from their initial releases (2.22.0, tagged 2026-06-03, and 3.2.0, tagged 2026-06-08).

marimo before 0.23.15 contains a configuration injection vulnerability that allows notebook authors to exfiltrate operator API keys by embedding a malicious base_url in PEP-723 inline script metadata, which is merged into session configuration with higher precedence than the operator's own settings due to insufficient sanitization in sanitize_pyproject_dict. When an operator opens the crafted notebook and makes an AI request, marimo resolves the attacker-controlled base_url from the notebook config while falling back to the operator's OPENAI_API_KEY environment variable for authentication, transmitting the API key to the attacker-controlled endpoint without requiring any cell execution.

Perspective 5.0.0 contains a denial of service vulnerability that allows remote attackers to block the server event loop indefinitely by submitting a crafted expression containing unbounded for or while loop constructs in a TableMakeViewReq message. Attackers can embed an arbitrarily large iteration count in an expression column evaluated once per table row, causing the Tornado IOLoop to block without any iteration cap, deadline, or cancellation check, rendering the server unresponsive to all connected clients.

Perspective 5.0.0 contains a remote code execution vulnerability that allows unauthenticated attackers to execute arbitrary operating system commands by submitting crafted expression strings to the PolarsVirtualServer backend, which passes client-supplied input directly to Python's eval() with only __builtins__={} cleared. Attackers can exploit Python object attribute traversal through the interpreter's loaded class list to reach subprocess.Popen via a TableValidateExprReq or TableMakeViewReq protobuf message, achieving arbitrary command execution in the Perspective host process.

The non-blocking (asynchronous) JSON parser in jackson-core does not enforce the maxNumberLength constraint defined in StreamReadConstraints (default: 1000 characters). An attacker able to submit JSON to an application that uses the async parser API can supply a number token of arbitrary length, leading to excessive memory allocation and potential CPU exhaustion, resulting in a denial of service. The synchronous parser enforces this limit correctly, so the constraint is applied inconsistently depending on which parsing API the application uses. Root cause: the async parsing path in NonBlockingUtf8JsonParserBase and related classes never invokes the number length validation methods. Number parsing methods such as _finishNumberIntegralPart() accumulate digits into the TextBuffer without any length check, then call _valueComplete() to finalize the token. _valueComplete() does not call resetInt() or resetFloat(), which are the methods in ParserBase where validateIntegerLength() and validateFPLength() are performed. Because that validation step is skipped, maxNumberLength is never enforced on the async code path. Impact: an attacker sending a JSON document containing an arbitrarily long number to an application using the async parser (for example a Spring WebFlux or other reactive application) can cause unbounded allocation in the TextBuffer and an OutOfMemoryError. If the application subsequently calls getBigIntegerValue() or getDecimalValue(), the JVM may additionally be tied up in O(n^2) BigInteger parsing, causing CPU-based denial of service. No privileges or user interaction beyond the ability to submit data for parsing are required. This issue affects com.fasterxml.jackson.core:jackson-core from version 2.15.0 through 2.18.5 and from 2.19.0 through 2.21.0, and tools.jackson.core:jackson-core from 3.0.0 through 3.0.x. Versions prior to 2.15.0 are not affected, because StreamReadConstraints -- which defines the maxNumberLength setting -- was first introduced in jackson-core 2.15.0, so no such constraint exists to be bypassed in earlier releases. Note that GHSA-72hv-8253-57qq records the lower bound of the affected 2.x range as 2.0.0.

Koha's reports/bor_issues_top.pl builds dynamic SQL in sub calculate by concatenating several user-controlled request parameters directly into the query string. The Criteria parameter is only normalized by a table-name prefix and is never whitelisted, landing verbatim in identifier positions (SELECT DISTINCTROW, GROUP BY, ORDER BY); Filter values are concatenated raw into single-quoted LIKE, BETWEEN, and comparison fragments, and the Limit parameter is appended raw to a LIMIT clause. An authenticated staff user holding the reports module permission can inject arbitrary SQL and read any table reachable by the Koha database user, including borrowers (password hashes, two-factor secrets, personal data), api_keys, and sessions.