vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a sanitize_message helper that strips object-repr memory addresses from error messages before they reach the client, is incomplete: several response paths echo str(exc) directly to clients without calling sanitize_message. The unsanitized sites include the Anthropic API router in vllm/entrypoints/anthropic/api_router.py (the POST /v1/messages and POST /v1/messages/count_tokens handlers), the Server-Sent Events streaming converter in vllm/entrypoints/anthropic/serving.py, and the realtime speech-to-text WebSocket in vllm/entrypoints/speech_to_text/realtime/connection.py. These paths catch the exception inside the route coroutine and construct the JSONResponse themselves, bypassing the sanitizing global FastAPI exception handler, and WebSocket frames do not traverse that handler chain at all. Using the same primitive as the parent issue, an unauthenticated attacker can send malformed image bytes through the Anthropic Messages API image content parts so that PIL.Image.open raises an UnidentifiedImageError whose message contains the BytesIO object repr, leaking the heap memory address verbatim in the error.message field of the response body. This vulnerability is fixed in 0.23.1rc0.

Python-Multipart is a streaming multipart parser for Python. Prior to 0.0.30, QuerystringParser 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 are tokenized into different fields than a WHATWG compliant intermediary would produce, allowing an attacker to smuggle extra form fields past an upstream body inspecting component. This vulnerability is fixed in 0.0.30.

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.1, the vLLM Dockerfile is vulnerable to a dependency confusion attack through the flashinfer-jit-cache package. The package is installed from a custom index (flashinfer.ai/whl/) using --extra-index-url, but the package name was not registered on PyPI, and UV_INDEX_STRATEGY="unsafe-best-match" is set globally. An attacker who registers flashinfer-jit-cache on PyPI with version 0.6.11.post2 can execute arbitrary code as root during the Docker build and backdoor every resulting container image, enabling exfiltration of all user prompts, API credentials, and model data from production vLLM deployments This vulnerability is fixed in 0.22.1.

The fix for CVE-2026-2443 was regressed by a subsequent rework commit that replaced specific overflow checks with a general signed comparison. When a client sends a Range request with a suffix length exceeding the content size, the resulting negative start value is not properly clamped, leading to malformed HTTP 206 responses and log flooding.

Python-Multipart is a streaming multipart parser for Python. Prior to 0.0.30, parse_options_header 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 filename*0/filename*1 continuation form) is decoded and surfaced under the bare filename/name key, and overrides the plain parameter when both are present. RFC 7578 §4.2 explicitly forbids the filename* form in multipart/form-data. Components that follow RFC 7578, or that do not implement RFC 2231/5987 decoding for multipart/form-data (WAFs, proxies, gateways), may interpret such a header differently. An attacker can exploit that difference to smuggle a different field name or filename past an upstream inspector to the backend. This vulnerability is fixed in 0.0.30.

Angular is a development platform for building mobile and desktop web applications using TypeScript/JavaScript and other languages. Prior to 22.0.0-rc.2, 21.2.15, 20.3.22 and 19.2.22, an issue in the @angular/compiler and @angular/core packages allows bypassing element and attribute sanitization/validation through specific namespace workarounds. Specifically, namespaced script elements (e.g., <svg:script> or <:svg:script>) were not properly identified as script elements by the Angular template preparser, allowing them to pass through template compilation without being stripped. Furthermore, security context schema mappings for element attributes did not consistently handle attributes within namespaced elements (like SVG and MathML), opening up gaps where malicious namespaced attributes could bypass runtime and compile-time sanitizers. Combined, these flaws enable an attacker who can inject or supply a template/tag structure with custom namespaces to bypass Angular's script-stripping logic and attribute sanitizers, leading to client-side Cross-Site Scripting (XSS). This vulnerability is fixed in 22.0.0-rc.2, 21.2.15, 20.3.22 and 19.2.22.

A heap-based buffer overflow was found in dnsmasq. When DNSSEC validation and query logging are both enabled, logging of DS or DNSKEY replies containing unsupported algorithm or digest types can cause dnsmasq to write past the end of an internal logging buffer. A remote attacker able to supply such a DNS response may crash the dnsmasq process, resulting in denial of service.

Angular is a development platform for building mobile and desktop web applications using TypeScript/JavaScript and other languages. Prior to 22.0.1, 21.2.17, and 20.3.25, a Denial of Service (DoS) vulnerability exists in the @angular/common package of the Angular framework. The formatDate function, which is also utilized by the standard Angular DatePipe, does not properly limit or validate the length of the format parameter. When parsing a maliciously crafted, excessively long date format string (e.g., a repeating pattern or very large string), the internal parser splits the string iteratively using a regular expression loop. This results in uncontrolled resource consumption (high CPU utilization and excessive memory allocations), leading to a Denial of Service (DoS). This vulnerability is fixed in 22.0.1, 21.2.17, and 20.3.25.

vLLM versions >= 0.6.3 and < 0.9.0 contain multiple regular expression denial of service (ReDoS) vulnerabilities. Several regex patterns — in vllm/lora/utils.py, the phi4mini tool parser, and the OpenAI-compatible serving chat endpoint — are susceptible to catastrophic backtracking. An attacker submitting crafted input with nested or repeated structures can trigger severe CPU consumption and performance degradation, resulting in denial of service.

Gophish through 0.12.1 contains a denial of service vulnerability that allows authenticated users with the User role to exhaust server memory by uploading a crafted Office document as an email template attachment. The ApplyTemplate() function in models/attachment.go processes Office documents as ZIP archives and calls ioutil.ReadAll() on each contained file entry without enforcing size restrictions on uncompressed content, allowing a zip bomb payload to expand to several gigabytes in memory and cause the process to be terminated by the operating system.