Intelligence
Explorer.

vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. Attackers can trigger GPU memory exhaustion by submitting completion requests with multiple prompts, causing orphaned KV cache blocks to accumulate until process restart and eventually preventing legitimate requests from executing.

vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate unbounded memory. Attackers can supply arbitrary tp_size values in prefill/decode disaggregated deployments to exhaust memory and trigger kernel OOM-kill of the decode worker process.

vLLM through 0.29.0 contains a resource exhaustion vulnerability in MooncakeConnector where rejected prefill requests create ownerless transfer placeholders that are never reclaimed. Attackers can send rejected requests to exhaust sender task pools, causing valid requests to be delayed by up to 480 seconds while health checks continue returning success.

vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with TieringOffloadingSpec and a peer-to-peer secondary tier. Attackers can supply arbitrary remote host and port values in kv_transfer_params to create unreachable peer sessions that retain ZeroMQ sockets until the context quota is exhausted, causing an uncaught ZMQError that crashes EngineCore and stops all inference.

vLLM through 0.29.0 contains a denial of service vulnerability in the NIXL connector's prefix caching implementation that fails to properly validate block counts across multi-prompt completion requests in prefill/decode disaggregated deployments. Attackers can trigger an assertion failure in NixlBaseConnectorWorker._apply_prefix_caching by submitting completion requests with multiple prompts of varying lengths, causing the decode worker to terminate and become unavailable until restarted.

vLLM versions through 0.29.0 contain a denial of service vulnerability in the NIXL connector's metadata handling for prefill/decode disaggregated deployments. Attackers can send requests with incomplete kv_transfer_params dictionary entries to trigger an uncaught KeyError in EngineCore scheduling, causing the decode engine to terminate and making all routed requests fail until manual restart.

DesktopSMS 1.11.0 by MrPear contains an unauthorized access vulnerability that allows local attackers to transmit SMS, retrieve SMS-derived content, and persist an attacker-selected paired identity by interacting with the application's local service without any pairing confirmation or user interaction. Attackers can exploit the unauthenticated local service through same-device loopback to perform privileged SMS operations using the victim application's permissions.

lamp-cloud through 5.10.0 fails to validate the employeeId parameter in the /anyone/visible/resource endpoint, allowing authenticated users to read any employee's roles and permissions. Attackers can supply arbitrary employeeId values to enumerate other employees' role codes, permission codes, and complete front-end router trees without authorization checks.

lamp-cloud through 5.10.0 contains an authorization bypass vulnerability in the deleteMyNotice endpoint that allows authenticated users to delete other users' notifications. Attackers can call the DELETE /anyone/extendNotice/deleteMyNotice endpoint with arbitrary notice IDs to permanently remove notifications belonging to other users without recipient validation.

lamp-cloud through 5.10.0 fails to validate user identity in PUT /anyone/baseInfo and PUT /anyone/avatar endpoints, allowing authenticated attackers to modify arbitrary user profiles. Attackers can supply target user IDs in request bodies to rewrite profile fields including nickname, ID card, sex, nation, education, work description, and avatar attachments of other users.

lamp-cloud through 5.10.0 contains an authorization bypass vulnerability in FileAnyoneController that allows authenticated users to download arbitrary attachments. Attackers can retrieve other users' stored files by supplying valid attachment identifiers to the /anyone/file/down and /anyone/file/download endpoints, as the application never validates file ownership against the created_by column.

lamp-cloud through 5.10.0 contains an authorization bypass vulnerability in the getUserInfoById endpoint that allows authenticated users to read any other user's full profile. Attackers can iterate the userId parameter to harvest sensitive user information including mobile numbers, email addresses, national identity card numbers, and WeChat and DingTalk OpenIDs.

Gladys Assistant before 5.1.0 contains a password reset link poisoning vulnerability that allows unauthenticated remote attackers to obtain valid password reset tokens for any account by exploiting the client-supplied origin parameter in the forgot_password endpoint without server-side validation. Attackers can send a crafted request specifying an attacker-controlled origin, causing the victim to receive a poisoned reset link that discloses the session token to the attacker, enabling full account takeover including administrator accounts.

Pagekit CMS <= 1.0.18 allows an unauthenticated attacker to perform SQL injection through the credentials array submitted to the public login endpoint (POST /user/authenticate).

Jazzware RT1000 Edge webUI v. 20.0.1 contains an unrestricted file upload vulnerability in the upgrade package upload functionality. An authenticated attacker can upload a server-side executable file. The uploaded file is stored in a web-accessible executable location and can be accessed directly over HTTP without authentication, resulting in remote code execution.

An issue in CrossWire Xiphos <= 4.3.2 allows a local attacker to execute arbitrary code via the src/main/url.cc and src/gtk/menu_popup.c components

An issue in gray-matter All versions (verified on 4.0.3) allows the JavaScript engine in lib/engines.js using eval() to parse front matter when language is js/javascript.This allows arbitrary code execution.

An issue in Matter Standard Specification-Implementation gap v1.5.1 Matter Project Chip V1.5.1 allows a local attacker to obtain sensitive information via the PerformCommissioningStep function in the ChipDeviceController.cpp component

Chartbrew is an open-source web application that can connect directly to databases and APIs and use the data to create charts. Prior to 5.2.3, Chartbrew's ClickHouse protocol in server/sources/plugins/clickhouse/clickhouse.protocol.js calls applySqlVariables() from server/sources/shared/sql/sql.variables.js without enabling the escapeBackslash option. For a ClickHouse-backed chart with variable binding, an attacker can supply a backslash before a quote so quote doubling does not keep the value within its intended SQL string literal. Public dashboards can expose this path without authentication, and successful exploitation can execute arbitrary ClickHouse SQL to disclose data or, when the database configuration permits, access files or internal network resources. This issue is fixed in version 5.2.3.

Chartbrew is an open-source web application that can connect directly to databases and APIs and use the data to create charts. Prior to 5.2.2, Chartbrew's runQuery() implementation in server/modules/ai/orchestrator/tools/runQuery.js interpolates the AI tool's row_limit parameter into a SQL LIMIT clause without runtime integer validation. The read-only keyword check runs before this value is appended, so an authenticated user who can influence a model-generated non-integer row_limit can add SQL that bypasses the earlier check. Successful exploitation can execute arbitrary statements against the connected database, including reading or changing data and, where database permissions permit, accessing files or executing operating-system commands. This issue is fixed in version 5.2.2.

Chartbrew is an open-source web application that can connect directly to databases and APIs and use the data to create charts. Prior to 5.2.2, Chartbrew's runQuery() implementation in server/modules/ai/orchestrator/tools/runQuery.js attempts to enforce read-only database access with a blocklist containing only seven SQL keywords. An authenticated user with AI feature access can submit dangerous statements or database functions that are absent from the read-only keyword blocklist, causing them to execute without SQL injection or keyword-obfuscation techniques. Depending on the database engine, configuration, and database-user privileges, this can expose or write files, access internal network resources, change database privileges, execute commands, or alter data. This issue is fixed in version 5.2.2.

Chartbrew is an open-source web application that can connect directly to databases and APIs and use the data to create charts. Prior to 5.2.2, Chartbrew's server/modules/safeRequest.js calls validateOutboundUrl() to resolve and validate a target hostname, but request-promise performs a separate DNS resolution for the actual connection. An authenticated user who can create or test API connections and controls the target DNS name can return a public address during validation and a private address during the request, bypassing the protections implemented by server/modules/outboundTargetPolicy.js. This DNS rebinding condition can expose internal services, localhost resources, or cloud metadata endpoints. This issue is fixed in version 5.2.2.

Zapros, a Python HTTP client, prior to version 0.14.0 is vulnerable to denial of service via memory exhaustion. The issue affects all callers who streamed compressed responses relying on the chunk size — explicit (`iter_bytes(chunk_size=...)`) or the default — to bound memory. The decoder ignored that bound, so a chunk could be far larger than requested and a single compressed response could overflow memory. Version 0.14.0 contains a patch. Some workarounds are available. Read the still-compressed body with `Response.iter_raw()` / `Response.async_iter_raw()`, which bypass the built-in decoders, and decompress it yourself with an explicit output-size bound (e.g. `zlib`'s `max_length`), aborting once a configured limit is exceeded. Where feasible, send `Accept-Encoding: identity` to disable response compression so bodies are not decompressed client-side. Avoid decoding response bodies from untrusted servers.

Zapros, a Python HTTP client, prior to version 0.14.0 is vulnerable to denial of service when an application requests content from an untrusted server, or follows a redirect to one, because a malicious response containing an excessive number of chained `Content-Encoding` values causes Zapros to construct a deeply nested decompression chain that consumes excessive resources. Version 0.14.0 patches the vulnerability by limiting responses to five content-encoding layers and raising `DecodingError` when that limit is exceeded. As a workaround, applications can add response middleware that inspects the `Content-Encoding` header and rejects responses containing more than a safe number of encoding layers.