Crawl4AI is an open-source LLM-friendly web crawler and scraper. Prior to 0.9.0, the Docker API server applied its SSRF destination check on the non-streaming /crawl path but not on the streaming path. handle_stream_crawl_request passed seed URLs straight to the crawler with no destination validation, allowing a remote unauthenticated client to call POST /crawl/stream or POST /crawl with crawler_config.stream=true with a URL pointing at an internal, private, or link-local address; the server fetched it and streamed the response body back. This issue is fixed in version 0.9.0.
Crawl4AI is an open-source LLM-friendly web crawler and scraper. Prior to 0.9.0, the Docker API server accepted request-supplied browser_config.extra_args, which flowed into Chromium's launch arguments. An attacker could inject Chromium switches that replace a child-process launch command together with --no-zygote, causing Chromium to fork or exec an attacker-controlled command as the container's runtime user. The Docker API is unauthenticated by default, so a single request yields arbitrary command execution. This issue is fixed in version 0.9.0.
Crawl4AI is an open-source LLM-friendly web crawler and scraper. Prior to 0.9.0, when the crawler saves a downloaded file, the destination filename was taken from attacker-influenced input and joined to the downloads directory with no confinement. A filename containing an absolute path or traversal escaped the downloads directory, giving an arbitrary file write with attacker-controlled contents; the HTTP crawler path uses the response Content-Disposition filename and the browser crawler path uses the download's suggested filename. Because the written bytes are attacker-controlled, this can escalate to remote code execution. This issue is fixed in version 0.9.0.
A flaw in the authentication mechanism for video stream requests in Genetec Security Center 5.14.0.0 prior to build 5.14.178.18 may allow an unauthenticated attacker to access live video streams.
vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Prior to 0.24.0, the structured_outputs.regex API parameter passes a user-supplied regular expression string directly to the grammar compiler backends with no compilation timeout; in the xgrammar backend the string reaches the regex compiler with no guard, and in the outlines backend the validation step blocks structural issues such as lookarounds and backreferences but performs no complexity analysis, so a pattern with nested quantifiers passes all checks and causes exponential state-space expansion, allowing a single request containing an adversarial regex to hang an inference worker indefinitely and deny service. This issue is fixed in version 0.24.0.
vLLM is a library for LLM inference and serving. From 0.12.0 to before 0.24.0, sending a pure prompt embeds payload in a /v1/completions request with a model using M-RoPE causes EngineCore to fail an assertion and fatally crash, shutting down the entire server application. Any remote user who is authorized to make a /v1/completions request can make such a request and induce a crash. This issue is fixed in version 0.24.0.
Traefik is an open source HTTP reverse proxy and load balancer. From v3.7.0 prior to v3.7.6, Traefik's Kubernetes Gateway API provider may resolve two accepted HTTPRoutes that target the same backend Service:port but configure different backendRef filters to the same child service and apply only one route's filter set to all requests reaching that backend. In Gateway deployments where backendRef filters set security-sensitive headers, such as tenant identity, authorization context, or values the backend trusts, an attacker who can create an accepted HTTPRoute sharing the same backend Service:port may cause their route's filter context to be applied to another route's requests, potentially crossing namespace boundaries when a ReferenceGrant permits cross-namespace targeting. This issue is fixed in version v3.7.6.
Traefik is an HTTP reverse proxy and load balancer. Prior to v2.11.51, v3.6.22, and v3.7.6, Traefik's ForwardAuth middleware, even when configured with trustForwardHeader: false, derives the X-Forwarded-Port header sent to the authentication service from the original incoming request instead of the sanitized forwarded request. As a result, an unauthenticated remote attacker can inject an X-Forwarded-Proto: https header over a plain HTTP connection and cause Traefik to forward X-Forwarded-Port: 443 to the authentication service, bypassing port-based authorization checks. This issue is fixed in versions v2.11.51, v3.6.22, and v3.7.6.
Traefik is an HTTP reverse proxy and load balancer. Prior to v2.11.51, v3.6.22, and v3.7.6, Traefik's BasicAuth, DigestAuth, and ForwardAuth middlewares strip canonical-cased spoofed identity headers before writing Traefik's own value, but do not account for underscore-variant header names, which many backends normalize identically to dashed forms. An attacker able to reach a protected route can inject an underscore-variant header that survives Traefik's stripping and reaches the backend alongside, or on the unauthenticated ForwardAuth authResponseHeaders path instead of, the value Traefik intended to set, spoofing identity or authorization context. This issue is fixed in versions v2.11.51, v3.6.22, and v3.7.6.
Rejected reason: ** REJECT ** DO NOT USE THIS CANDIDATE NUMBER. ConsultIDs: CVE-2026-54637. Reason: This candidate is a duplicate of CVE-2026-54637. Notes: All CVE users should reference CVE-2026-54637 instead of this candidate.
vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Prior to 0.24.0, a frontend-legal multi-request speculative decoding workload can cause the rejection sampler to produce a recovered token equal to the model vocabulary size boundary value, which is then converted to negative one when the engine selects the next live token for a request and is written back into the drafter's input ids; that out-of-vocabulary value is later consumed by the model's embedding and attention path and crashes the engine worker with a GPU device-side assertion. The same triggering request sequence is reachable through the public gRPC Generate and Abort endpoints, so a remote client that can send generation requests can crash the shared engine worker, aborting concurrent requests and causing a service-wide denial of service for other clients of the deployment until the worker is restarted. This issue is fixed in version 0.24.0.
Hugo is a static site generator. From 0.123.0 to 0.161.1, a regression made RootMappingFs.statRoot use Stat (follows symlinks) instead of Lstat , so a direct resources.Get of a symlink pointing outside its mount returned the target's contents — letting a symlink planted in a local mount (e.g. a vendored themes/ theme) read arbitrary files accessible to the Hugo user. Go-module themes from GitHub (symlinks stripped) and directory walks were unaffected. Fixed in 0.162.0.
DNG SDK versions 1.7.1 2536 and earlier are affected by a NULL Pointer Dereference vulnerability that could result in an application denial-of-service. An attacker could exploit this vulnerability to crash the application, leading to a denial-of-service condition. Exploitation of this issue requires user interaction in that a victim must open a malicious file.
FOSSBilling is a free, open-source billing and client management system. Versions 0.6.0 through 0.7.2 have an unauthenticated payment bypass vulnerability in FOSSBilling's IPN callback endpoint. When the Custom payment adapter is enabled, an attacker can mark any unpaid invoice as paid and credit the associated client account without making an actual payment, by sending a single crafted HTTP request. Version 0.8.0 patches the issue. Some workarounds are available. Disable the Custom payment gateway if not actively needed and/or restrict access to `/ipn.php` at the web server level (e.g., via IP allowlisting), noting that this may interfere with legitimate payment callback processing.
FOSSBilling is a free, open-source billing and client management system. Prior to version 0.8.0, the Guest API invoice/update endpoint is missing an authorization check present in other invoice-related endpoints, allowing an unauthenticated user with knowledge of an invoice hash to modify the payment gateway associated with an unpaid invoice. An attacker who obtains an invoice hash, which may leak through shared URLs, referrer headers, or email links, can change the `gateway_id` on an unpaid invoice to any payment gateway configured in the system. This does not allow redirecting payments to an arbitrary external endpoint, as the gateway must already be installed and configured by an administrator. The practical impact is further limited by the `invoice_accessible_from_hash` system setting. Version 0.8.0 contains a patch. No known workarounds are available.
Coolify is an open-source and self-hostable tool for managing servers, applications, and databases. Prior to 4.0.0-beta.469, an authenticated remote command injection vulnerability in application deployment handling allows users with application write permissions to achieve remote code execution and exfiltrate sensitive environment variables through deployment logs via fields such as dockerfile_location and deployment commands. This issue is fixed in version 4.0.0-beta.469.
FOSSBilling is a free, open-source billing and client management system. Versions 0.6.0 through 0.7.2 have a SQL injection vulnerability in the `Massmailer` module filter functionality. An authenticated administrator can supply crafted filter values when updating a mass email message, causing untrusted input to be interpolated directly into SQL in the recipient selection query. Version 0.8.0 patches the issue. Some workarounds are available. Restrict administrator access to trusted users only, disable the `Massmailer` module if it is not required, audit existing records in the `mod_massmailer` table for suspicious filter values, and/or review administrator activity related to `Massmailer` message updates.
Cryptographic Issue when using a static initialization vector for AES-GCM key wrapping, which requires a unique value for each call to ensure security.
Improper Neutralization of Special Elements in the metrics-service retention policy management component in Amazon mcp-gateway-registry before 1.0.13 might allow an authenticated remote user to execute arbitrary SQL queries via a crafted table_name value that is interpolated into SQL statements in identifier position.
To remediate this issue, users should upgrade to version 1.0.13 or later.
HashiCorp Terraform Enterprise contained an issue in its version control system (VCS) ingestion of registry modules that did not correctly enforce the intended boundary on packaged module content. This may allow an authenticated user to include files from outside the intended repository content in a module and then download them, potentially exposing sensitive files readable by the ingestion process. This vulnerability, CVE-2026-14468, is fixed in Terraform Enterprise v2.0.4 and v1.2.4.
Memory Corruption when invoking device input/output control operations for mapping and unmapping persistent memory buffers due to improper synchronization.
A flaw was found in GIMP. The PlayStation TIM loader, responsible for handling PlayStation image files, incorrectly calculates the size of the Color Look-Up Table (CLUT) due to an integer overflow. This occurs when multiplying num_colors and num_cluts, both 16-bit unsigned short integers, resulting in a value exceeding the maximum integer limit. An attacker could exploit this by providing a specially crafted image file, leading to undefined behavior and causing the GIMP plug-in to abort, effectively resulting in a denial of service.
Hugo is a static site generator. From v0.162.0 through v0.163.0, the default security.http.urls policy denies requests to loopback, internal, and cloud-metadata IPv4 literals, but the deny rule only matched dotted-decimal notation, so alternate IPv4 encodings of the same addresses, including integer, hex, or octal, passed the policy. When a template passes an untrusted or data-derived URL to resources.GetRemote and the host platform uses the cgo system resolver, these encodings resolve to the blocked address, allowing build-time server-side requests to loopback and internal services, including the cloud-metadata endpoint in hosted or CI builds; the same check is reused on redirects, so the gap also applies to each redirect hop. This issue is fixed in v0.163.1.
Hugo is a static site generator. From v0.123.0 through v0.163.0, Hugo's virtual filesystem is designed so that files under a mount cannot reach outside the mount tree, but a regression caused RootMappingFs.statRoot to call Stat, which follows symlinks, instead of Lstat, so a direct os.ReadFile "somefile" where somefile was a symlink pointing outside the mount would return the target's contents. This effectively let a symlink planted inside a theme or local mount read arbitrary files reachable to the user running hugo. This issue is fixed in v0.163.1.
Hugo is a static site generator. From 0.60.0 until 0.163.3, Hugo's default code-block renderer wrote the Markdown code-fence language or info-string into the code class="language-…" data-lang="…" wrapper without HTML escaping. A fence info-string containing a quote and a script payload breaks out of the attribute and injects a live script element. This issue is fixed in 0.163.3.
vLLM is an inference and serving engine for large language models. From 0.22.0 to 0.23.0, the /v1/audio/transcriptions and /v1/audio/translations routes call request.file.read() to fully materialize an uploaded audio file into memory before vLLM checks the documented VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed upload size limit (default 25 MB) later in the speech-to-text preprocessing step, so an API caller who can reach those routes can submit an oversized multipart upload and cause vLLM to allocate memory proportional to the uploaded file size before the request is rejected as too large, creating memory pressure or terminating the process depending on deployment resource limits. This issue is fixed in version 0.24.0.
OP-TEE is a Trusted Execution Environment (TEE) designed as companion to a non-secure Linux kernel running on Arm; Cortex-A cores using the TrustZone technology. Starting in version 3.0.0 and prior to version 4.11.0, 32-bit integer overflows in OP-TEE core's AES-GCM implementation cause the authentication tag to be computed with incorrect bit-length values after processing more than 512 megabytes of payload or Additional Authenticated Data (AAD). Version 4.11.0 contains a patch. No known workarounds are available.
Hugo is a static site generator. From 0.91.0 until 0.162.0, resources.GetRemote enforces security.http.urls on the URL it is called with, but it did not re-validate intermediate URLs on HTTP 3xx redirects. An allowed server (or an attacker controlling its DNS or response) could therefore redirect the request to a host that the policy was meant to forbid and Hugo would fetch from the redirected target. The same bypass also lifted any host-shape restriction the operator had put in place. This vulnerability is fixed in 0.162.0.
Hugo is a static site generator. Prior to 0.162.0, Hugo accepts content files in several markup formats. Files mapped to the text/html media type (typically .html files under /content, or pages produced by a content adapter that sets content.mediaType = "text/html") had their body emitted verbatim into the rendered page. A site that ingests HTML content from an untrusted source could therefore be served stored cross-site scripting. This vulnerability is fixed in 0.162.0.
OP-TEE is a Trusted Execution Environment (TEE) designed as companion to a non-secure Linux kernel running on Arm; Cortex-A cores using the TrustZone technology. Starting in version 3.20.0 and prior to version 4.11.0, a vulnerability in OP-TEE’s subkey rollback protection allows the use of revoked or older subkey versions because the system fails to propagate versioning data during the Trusted Application (TA) loading process. In `core/crypto/signed_hdr.c`, the function `shdr_load_pub_key()` parses subkey headers but does not assign the `subkey_version` to the runtime `shdr_pub_key` structure. As a result, the `key->version` field remains at zero regardless of the version specified in the header. When `ree_fs_ta_open()` in `core/kernel/ree_fs_ta.c` calls `check_update_version()`, it passes this zeroed version to the rollback database. Because the database never receives a non-zero version to record, it never advances, effectively bypassing the rollback check and allowing TAs signed with downgraded subkey chains to load successfully. This impacts OP-TEE mainline configurations that utilize subkey-based signing chains for Trusted Application (TA) authentication. Version 4.11.0 contains a patch. No known workarounds are available.
OP-TEE is a Trusted Execution Environment (TEE) designed as companion to a non-secure Linux kernel running on Arm; Cortex-A cores using the TrustZone technology. Starting in version 3.3.0 and prior to version 4.11.0, a resource leak exists in OP-TEE’s shared memory cleanup logic because the function `cleanup_shm_refs()` in `core/tee/entry_std.c` fails to apply a required bitmask (`OPTEE_MSG_ATTR_TYPE_MASK`) to parameter attributes. When processing non-contiguous memory parameters from a normal-world caller, the system fails to match the attribute type in its internal switch statement and skips the necessary mobj_put() call. This results in a persistent reference leak of `mobj_reg_shm` objects, which remain on internal lists with dangling refcounts. This affects non-FF-A configurations that support non-contiguous, non-secure shared memory. Over time, these accumulated leaks progressively consume the secure-world heap, degrading the system's ability to service trusted application operations and eventually requiring a reboot to recover. Version 4.11.0 contains a patch. No known workarounds are available.
OP-TEE is a Trusted Execution Environment (TEE) designed as companion to a non-secure Linux kernel running on Arm; Cortex-A cores using the TrustZone technology. Starting in version 4.5.0 and prior to version 4.11.0, the RSA PKCS#1 v1.5 decryption implementation in the Hisilicon HPRE crypto driver uses non-constant-time `memcmp()` for label hash verification and has multiple distinguishable error paths. This creates a Bleichenbacher-style padding oracle that allows an attacker to recover RSA PKCS#1 v1.5 plaintext. Version 4.11.0 contains a patch. As a workaround, disable Hisilicon HPRE RSA driver with `CFG_HISILICON_ACC_V3=n`.
OP-TEE is a Trusted Execution Environment (TEE) designed as companion to a non-secure Linux kernel running on Arm; Cortex-A cores using the TrustZone technology. Starting in version 3.9.0 and prior to version 4.11.0, the RSA-OAEP decryption implementation in the NXP CAAM crypto driver uses non-constant-time `memcmp()` for label hash verification and has multiple distinguishable error paths. This creates a Manger-style padding oracle that allows an attacker to recover RSA-OAEP plaintext with approximately 1000-2000 adaptive chosen ciphertext queries. Version 4.11.0 contains a patch. As a workaround, disable the NXP CAAM RSA driver with `CFG_CRYPTO_DRV_RSA=n`.
OP-TEE is a Trusted Execution Environment (TEE) designed as companion to a non-secure Linux kernel running on Arm; Cortex-A cores using the TrustZone technology. Starting in version 4.5.0 and prior to version 4.11.0, the RSA-OAEP decryption implementation in the Hisilicon HPRE crypto driver uses non-constant-time `memcmp()` for label hash verification and has multiple distinguishable error paths. This creates a Manger-style padding oracle that allows an attacker to recover RSA-OAEP plaintext with approximately 1000-2000 adaptive chosen ciphertext queries. Only affects plat-d06 with `CFG_HISILICON_ACC_V3=y`, which seems to be disabled by default. Version 4.11.0 contains a patch. As a workaround, disable Hisilicon HPRE RSA driver with `CFG_HISILICON_ACC_V3=n`.
The OpenAI Codex desktop app for macOS rendered remote images from Markdown in model responses. An attacker who could place an indirect prompt injection in content processed by Codex, such as a connected-tool result or another untrusted source, could induce the model to construct a remote image URL containing sensitive data. The app automatically fetched that URL when rendering the response, sending the embedded data to an attacker-controlled server without a separate user click. Successful exploitation could exfiltrate secrets and other information accessible in the Codex session, including API keys, source code, and data returned by connected tools. No direct integrity or availability impact was demonstrated, and there is no known exploitation in the wild.
Improper enforcement of a mandatory multi-factor authentication policy in Devolutions Server 2026.2.9.0 allows an attacker with valid user credentials to bypass the MFA Required policy and authenticate without completing multi-factor authentication. The problem occurs when DVLS encounters an invalid default MFA value.
The web server binary /bin/httpd contains a hidden backdoor authentication mechanism in the login() function at 004c88b8.
- The function contains a normal authentication path using MD5/hash-based password verification (prod_encode64/PasswordToMd5/check_rand_key).
- After normal authentication fails, it calls GetValue("sys.rzadmin.password") to read a backdoor password from the device configuration.
- It performs a direct strcmp() comparison (plaintext, not hashed) between the config value and the user-supplied password.
A successful match grants role=2 (admin-level access) and creates a valid session. The rzadmin username is never checked — any username works with the backdoor
ArcGIS Server contains an unrestricted file upload vulnerability. An unauthenticated attacker could exploit this issue by uploading a crafted file to the affected endpoint. Successful exploitation could allow arbitrary file upload.
ArcGIS Server contains a directory traversal vulnerability. An unauthenticated attacker could exploit this issue by sending crafted path parameters. Successful exploitation could allow access to sensitive files on the system. This issue impacts all versions of ArcGIS Server 12.0 and prior.
Pillow is a Python imaging library. Prior to 12.3.0, WindowsViewer.get_command() constructed a cmd.exe shell command by directly embedding a file path into an f-string without escaping and passed the result to subprocess.Popen(..., shell=True), allowing shell metacharacters in the file path to inject arbitrary cmd.exe commands. This issue is fixed in version 12.3.0.
Pillow is a Python imaging library. Prior to 12.3.0, PIL/GdImageFile.py GdImageFile._open() read image dimensions from the GD 2.x header and stored them in self._size without calling Image._decompression_bomb_check(), allowing a crafted .gd file to trigger excessive C-heap allocation when loaded. This issue is fixed in version 12.3.0.