| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| USB Test is a lightweight Windows CLI utility for verifying sector integrity and detecting overreported capacity on USB flash drives. Prior to version 1.1.5, getSectorSize() in USB Test/src/driveinfo.cpp trusts DISK_GEOMETRY.BytesPerSector returned by IOCTL_DISK_GET_DRIVE_GEOMETRY without checking whether the call succeeded or enforcing size bounds. USB Test/src/drivetest.cpp uses the value for division and to size a heap buffer before memcpy(buffer.data(), §or, 8), so a physically connected malicious USB storage device can report a value below eight bytes and cause repeated out-of-bounds writes, or report zero and cause a divide-by-zero denial of service when a user starts a drive test. The heap corruption can crash the elevated application and may permit arbitrary code execution. This issue is fixed in version 1.1.5. |
| An out-of-bounds write issue was addressed with improved bounds checking. This issue is fixed in macOS Golden Gate 27.2, macOS Tahoe 26.7.2. Mounting a maliciously crafted disk image may cause unexpected system termination or corrupt kernel memory. |
| vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can provide a template that is compiled and rendered by tokenizer.apply_chat_template() without a wall-clock timeout, output-size limit, or iteration bound. Rendering runs in a ThreadPoolExecutor with one worker by default, so a nonterminating render can block /v1/responses and other chat-template users after the request ends, while an output-expanding render can exhaust server memory. Harmony models are not affected because they do not use this Jinja path. This issue is fixed in version 0.32.0. |
| vLLM is an inference and serving engine for large language models. Prior to 0.32.0, an authenticated caller can place forged shared-memory address and monotonic_id values in public multimodal features.kwargs_data. In scale-out deployments with shared-memory multimodal caching enabled, workers trust those values and pass attacker-selected bytes to pickle.loads(), allowing arbitrary Python execution even when insecure serialization is disabled. This issue is fixed in version 0.32.0. |
| SimulationCraft is a tool to explore combat mechanics in World of Warcraft. In builds after release-820-01 and before the identified May 7, 2026 nightly, the `bcp_api.cpp` loop over the JSON `sockets` array uses `slot_data["sockets"].Size()` without bounding `gem_idx` by `MAX_GEM_SLOTS`, which is four in `engine/sc_enums.hpp`. The destination `gem_id` and `gem_bonus_id` arrays in `engine/item/item.hpp` have only four elements. More than four socket entries therefore write beyond `gem_id` into the adjacent `gem_bonus_id` vector storage; `bonus_list` entries can also access `gem_bonus_id` out of bounds. Corrupting a vector's `_M_start` pointer causes the `item_t` destructor to free an attacker-influenced pointer during normal cleanup. The advisory reports crashes and code execution demonstrated in a testing environment. Attack inputs can arrive through a malicious `.simc` profile with a `[[equipment]]` section, a compromised addon export, or `Battle.net API` responses modified by an attacker with a privileged network position. The fix is contained in the nightly build identified as simc-1205-01-516b5d7; no containing release tag is available. |
| vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the Phi-4 multimodal audio encoder in phi4mm_audio.py and phi4mm_utils.py uses eager attention by default, allowing a remote client to submit one legal long audio clip that causes MultiHeadedAttention.forward_attention to materialize and retain full attention matrices in self.attn. The scaled dot-product attention branch does not materialize these matrices. The resulting quadratic GPU-memory use can cause an out-of-memory condition and terminate EngineCore for all users. This issue is fixed in version 0.32.0. |
| vLLM is an inference and serving engine for large language models. From 0.11.1 until 0.32.0, the DeepSeek-OCR, DeepSeek-OCR-2, and Unlimited-OCR image path allows a remote client that can reach the OpenAI-compatible API to submit a legal image whose normalized pixels sum to zero when that image is alone in its per-step encoder batch. _parse_and_validate_image_input returns no embedding, embed_multimodal propagates that value, and sanity_check_mm_encoder_outputs triggers a fatal assertion that shuts down the engine for all users. This issue is fixed in version 0.32.0. |
| A buffer overflow in RandR
ChangeProviderProperty requests in X.Org X Server before 2.1.25 and XWayland before 24.4.14 could be used by authenticated X clients to crash or execute code within the X server context. |
| vLLM is an inference and serving engine for large language models. Prior to 0.32.0, ImageMediaIO passes attacker-controlled image bytes to Pillow through Image.open(BytesIO(data)) without restricting decoder selection, and the MIME label does not constrain the selected format. Image-capable request paths through parse_image() and fetch_image() can therefore dispatch an EPS payload to host components. A host with Ghostscript affected by CVE-2024-29510 can execute commands with the vLLM process's privileges, while environments using Pillow 12.0.0 through 12.2.0 can trigger CVE-2026-59203 and exhaust media-loading workers. Deployments without the applicable vulnerable dependency are not affected by that corresponding impact. This issue is fixed in version 0.32.0. |
| YosemiteCrew provides organization-based service and task management. In the source-reported affected main and dev snapshots, YosemiteCrew's service and PMS task-library/template routes omit organization and permission checks. `authorizeCognito` in `apps/backend/src/middlewares/auth.ts` sets the authenticated user's ID but does not provide the `withOrgPermissions()` and `requirePermission(...)` checks used by protected siblings. In `apps/backend/src/routers/service.router.ts`, `PATCH /fhir/v1/service/:id` and `DELETE /fhir/v1/service/:id` are authentication-only, while `GET /fhir/v1/service/:id` has no authentication middleware. The trace through `apps/backend/src/controllers/web/service.controller.ts` to `apps/backend/src/services/service.service.ts` passes a global record ID into `ServiceModel.findById(oid)` without an `organisationId` filter. In `apps/backend/src/routers/task.router.ts`, `PUT /v1/task/pms/library/:libraryId`, `POST /v1/task/pms/templates`, `PATCH /v1/task/pms/templates/:templateId` and `DELETE /v1/task/pms/templates/:templateId` likewise lack organization/role enforcement. Any registered platform user, including a patient/companion-app account, can modify or delete another organization's service catalogue and PMS task-library/template data, affecting bookings and billing; service records can additionally be read without authentication. An earlier security change protected create routes but left these sibling operations exposed. A fixed release cannot be established from the available evidence. |
| vLLM is an inference and serving engine for large language models. Prior to 0.32.0, vllm/tokenizers/hf.py computes max_token_id with an off-by-one upper bound, allowing a pre-tokenized prompt ID equal to the vocabulary size to pass validation. For a tensor-parallel size of one and an exact multiple-of-64 vocabulary, VocabParallelEmbedding.forward does not mask that value before F.embedding uses it, causing a CUDA device-side assertion and terminating the engine. This issue is fixed in version 0.32.0. |
| vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the OpenAI-compatible server configures Starlette CORSMiddleware with allowed_origins set to a wildcard by default. A malicious web page can use a victim's browser to reach an unauthenticated vLLM server on localhost, an internal network, or a VPN, call endpoints including GET /v1/models and POST /v1/chat/completions, consume GPU resources, and read responses. Cookie-based cross-site request forgery does not apply because credentials are disabled by default, and deployments requiring bearer authentication are exposed only when the caller already has a usable token. This issue is fixed in version 0.32.0. |
| vLLM is an inference and serving engine for large language models. From 0.16.0 until 0.32.0, the /v1/responses endpoint allows an authenticated caller to place an attacker-controlled template in chat_template_kwargs[chat_template] when no server template is configured. ChatParams.get_apply_chat_template_kwargs() preserves that value without enforcing trust_request_chat_template, so the request can bypass the intended trust policy and cause CPU or memory exhaustion through Jinja rendering. This issue is fixed in version 0.32.0. |
| A logic issue was addressed with improved checks. This issue is fixed in macOS Golden Gate 27.2, macOS Sequoia 15.8.2, macOS Tahoe 26.7.2. A local attacker may be able to elevate their privileges. |
| Unverified ownership of the default query result location in the Athena module in AWS SDK for pandas (awswrangler) before 3.16.1 might allow remote actors to read Athena query results and tamper with result sets returned to the caller via a pre-created S3 bucket.
To remediate this issue, users should upgrade to version 3.16.1 or later. Users should also verify that the default Athena result bucket in each account and region is owned by their own account, and set an explicit result location or enforce a workgroup result configuration. |
| Xibo is an open source digital signage platform with a web content management system and Windows display player software. Prior to 4.4.3, missing authorization on `POST /displaygroup/criteria/{displayGroupId}` allows an authenticated user to send criteria updates to any display via `XMR`. These criteria determine schedule changes and can alter playback, for example by causing content scheduled for rainy weather to play when it is not raining. Exploitation requires `Schedule Criteria` to be in use on the CMS. This issue is fixed in version 4.4.3. |
| Photoshop Desktop is affected by an out-of-bounds write vulnerability that could result in arbitrary code execution in the context of the current user. Exploitation of this issue requires user interaction in that a victim must open a malicious file. |
| GFI KerioConnect (Kerio Connect) Outlook Connector (Offline Edition), 'KOFF', before 10.0.9 Patch 2 contains a local privilege escalation vulnerability caused by weak installer permissions that leave the product directory writable by the built-in Users group. A local non-administrative attacker can replace the searchd.exe service binary used by the bundled KOFFIndexingService, which runs as LocalSystem, to execute arbitrary code with NT AUTHORITY\SYSTEM privileges upon the next service start or reboot. |
| An out-of-bounds write issue was addressed with improved bounds checking. This issue is fixed in macOS Golden Gate 27, macOS Sequoia 15.8.2, macOS Tahoe 26.7.2. Processing a maliciously crafted file may lead to arbitrary code execution. |
| Vulnerability in the Siebel CRM Integration product of Oracle Siebel CRM (component: EAI). Supported versions that are affected are 17.0-26.9. Difficult to exploit vulnerability allows unauthenticated attacker with network access via HTTP to compromise Siebel CRM Integration. Successful attacks require human interaction from a person other than the attacker and while the vulnerability is in Siebel CRM Integration, attacks may significantly impact additional products (scope change). Successful attacks of this vulnerability can result in unauthorized access to critical data or complete access to all Siebel CRM Integration accessible data as well as unauthorized update, insert or delete access to some of Siebel CRM Integration accessible data. CVSS 3.1 Base Score 6.9 (Confidentiality and Integrity impacts). CVSS Vector: (CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:C/C:H/I:L/A:N). |
