| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| atomic-agents-stack before 1.1.0 contains a cost-guardrail bypass in the _estimate_batch_cost function that returns zero cost for unknown models not in the pricing table. Attackers can configure deployments with unknown model identifiers to bypass daily cost caps and exceed budget limits in parallel batch operations. |
| A security flaw has been discovered in vllm-project vLLM up to 0.29.0. The affected element is the function TiktokenTokenizer::new of the file rust/src/text/src/backend/hf/mod.rs of the component tiktoken vocab File Handler. The manipulation results in denial of service. The attack is only possible with local access. The exploit has been released to the public and may be used for attacks. The pull request to fix this issue awaits acceptance. |
| In initForUserNoTracing of VoiceInteractionManagerService.java, there is a possible persistent denial of service due to resource exhaustion. This could lead to local denial of service with no additional execution privileges needed. User interaction is not needed for exploitation. |
| Affected versions of MISP do not consistently apply the existing authentication-failure logging throttle.
Two API authentication failure branches wrote directly to the Log model:
- API requests with no authentication key;
- requests supplying an API key with an incorrect length
Unlike other authentication failures, these paths bypassed _shouldLog(), so every request could create another durable auth_fail entry.
Version affected: ≤2.5.45 |
| Hoverfly is an open source API simulation tool. Prior to version 1.12.8, remote post-serve actions use `http.DefaultClient` without any timeout configuration. When the remote endpoint is unreachable or intentionally slow (accepts TCP connection but never responds), each triggered proxy request spawns a goroutine that blocks indefinitely on `http.DefaultClient.Do()`. An attacker can cause unbounded goroutine accumulation leading to memory exhaustion and process crash (OOM kill). Unlike local post-serve action execution, this requires no binary execution, only a URL pointing to a non-responsive endpoint. Version 1.12.8 patches the issue. |
| webtransport-go is an implementation of the WebTransport protocol. Prior to 0.11.1, Session.parseNextCapsule() in session.go skips an unknown WebTransport capsule on the HTTP/3 request stream by calling io.ReadAll on the capsule reader, retaining the complete declared capsule body in memory. A malicious peer can send a large unknown capsule and cause a client or server to allocate memory for the full payload; QUIC flow control does not bound the total retained allocation because reading advances the flow-control window while the received bytes remain in memory. The resulting memory and resource exhaustion can disrupt or crash the affected process. This issue is fixed in version 0.11.1. |
| Mattermost versions 11.9.x <= 11.9.0, 11.8.x <= 11.8.4, 11.7.x <= 11.7.7, 10.11.x <= 10.11.22 fail to properly enforce the limit of concurrent files being processed and handled failed files, which allows a user with permission to upload files to spawn more goroutines than intended and block the indexing of other files via uploading heavy files constantly to the server.. Mattermost Advisory ID: MMSA-2026-00696 |
| Description
The DRPC server kept a map from function name to request queue and created an entry the first time a
function name was seen. No code path ever removed an entry: request cleanup removed the request from its
queue, and the shutdown path drained queues, but the queue object and its map entry remained for the life of
the process.
Function names come from the client and are not constrained to functions any topology has registered, so the
number of retained entries is bounded only by the number of distinct names an attacker chooses to send, and
each retained entry holds the name itself. `drpc.authorizer` is unset by default, so no credentials are
required to reach the endpoint.
The retained state is permanent rather than a transient load spike, so the effect accumulates until the DRPC
server exhausts its heap.
Mitigation
Upgrade to 3.1.0, where a function's queue is removed once nothing is waiting in it.
Users who cannot upgrade immediately should configure `drpc.authorizer` so that only trusted principals can
reach the DRPC endpoints, and should ensure the DRPC ports are not reachable from untrusted networks.
Credit
The ASF -- found using Claude agents to study the security of open-source projects, validated and reported by Apache Storm. |
| dd-trace-java is a Datadog APM client for Java. Prior to 1.62.0, W3C baggage extraction does not enforce DD_TRACE_BAGGAGE_MAX_ITEMS, which defaults to 64, or DD_TRACE_BAGGAGE_MAX_BYTES, which defaults to 8192, although those limits apply during baggage injection. A remote unauthenticated attacker can send a baggage HTTP header containing many comma-separated key-value pairs or a single very large value. The extraction path allocates map entries while parsing the attacker-controlled header on every request, causing unbounded CPU and memory consumption in an HTTP service where the baggage propagation style is enabled, which is the default for most affected tracers. This can cause denial of service. This issue is fixed in version 1.62.0. |
| OpenTelemetry eBPF Profiler is a production-scale agent for profiling applications across multiple programming languages. Starting in version 0.0.202527 and prior to version 0.0.202622, an unprivileged process can cause the profiler to open a nonregular mapping file, such as a FIFO, and block indefinitely, preventing further ELF analysis and causing a denial of service. Version 0.0.202622 contains a patch. No known workarounds are available. |
| Traefik is an open source HTTP reverse proxy and load balancer. From 2.8.2 until 2.11.56 and 3.7.12, HTTP/3 entrypoints do not apply entryPoints..transport.respondingTimeouts.readTimeout because the timeout is enforced on a TCP connection and the HTTP/3 server has no corresponding QUIC stream deadline. An unauthenticated client can use a slow request body, trickling data indefinitely while holding a request and an upstream connection open and exhausting backends with bounded connection pools. This issue is fixed in 2.11.56 and 3.7.12. |
| webhook through 2.8.3 reads the entire request body into memory before evaluating trigger rules, allowing unauthenticated attackers to exhaust memory by sending oversized bodies. Attackers can send multi-gigabyte request bodies with invalid signatures to trigger out-of-memory conditions and crash the service. |
| vLLM versions >=0.10.2 and <0.28.0 do not apply any audio decode-size or duration limit when extracting audio from video input for NanoNemotronVL models. In nano_nemotron_vl.py, _extract_audio_from_videos calls load_audio_pyav(BytesIO(video_bytes)) without the max_duration_s or max_decode_bytes parameters, so neither VLLM_MAX_AUDIO_DECODE_DURATION_S nor VLLM_MAX_AUDIO_DECODE_BYTES is enforced (unlike the direct audio upload path in AudioMediaIO). When a NanoNemotronVL model is served with use_audio_in_video=True, an attacker who supplies a small, highly compressed video as multimodal input can force the server to allocate gigabytes of memory during audio decoding, resulting in a denial of service. Fixed in vLLM 0.28.0. |
| AirSane is a SANE frontend, and a scanner server that supports Apple's AirScan protocol. Versions prior to 0.4.12 have a vulnerability in the custom HTTP server implementation of AirSane that allows a remote unauthenticated attacker to cause a Denial of Service (DoS) via memory exhaustion (OOM). In httpserver.cpp, the HttpServer::Request::content function reads the Content-Length header and directly passes this value to std::string::resize() without any upper-bound validation or safe parsing. An attacker can send an HTTP POST request with an artificially large Content-Length value. This forces the daemon to attempt allocating gigabytes of memory, resulting in a std::bad_alloc exception and immediately crashing the AirSane process. Additionally, providing non-numeric characters in the Content-Length header leads to undefined behavior (NaN to integer conversion) due to the lack of error handling during header parsing. Version 0.4.12 patches the issue. |
| dd-trace-rb is Datadog's client library for Ruby. Prior to 2.32.0, W3C baggage extraction does not enforce DD_TRACE_BAGGAGE_MAX_ITEMS, which defaults to 64, or DD_TRACE_BAGGAGE_MAX_BYTES, which defaults to 8192, although those limits apply during baggage injection. A remote unauthenticated attacker can send a baggage HTTP header containing many comma-separated key-value pairs or a single very large value. The extraction path allocates entries while parsing the attacker-controlled header on every request, causing unbounded CPU and memory consumption in an HTTP service where the baggage propagation style is enabled, which is the default for most affected tracers. This can cause denial of service. This issue is fixed in version 2.32.0. |
| A weakness has been identified in TooTallNate Java-WebSocket up to 1.6.1. The impacted element is the function processFrameContinuousAndNonFin of the file Draft_6455.java of the component Fragmentation Handler. Executing a manipulation can lead to allocation of resources. The attack may be performed from remote. The exploit has been made available to the public and could be used for attacks. The pull request to fix this issue awaits acceptance. |
| vLLM versions before 0.28.0 fail to validate audio sample rate headers in the transcription endpoint, allowing authenticated clients to bypass duration checks. Attackers can submit forged FLAC headers with inflated sample rates to trigger excessive memory allocation and crash the API server process affecting all tenants. |
| node-opcua is an OPC UA implementation for TypeScript and Node.js. Prior to 2.166.0, the process-global g_alreadyUsedNonce cache used by nonceAlreadyBeenUsed in packages/node-opcua-secure-channel/source/server/server_secure_channel_layer.ts records nonces from OpenSecureChannelRequest and CreateSession without expiration or a size limit. An unauthenticated remote attacker can repeatedly create sessions with unique nonces, causing entries to persist after session expiry and accumulate across connection cycles even when maxSessions=10 limits concurrent sessions. The resulting unbounded heap growth can exhaust the default Node.js heap and crash the node-opcua server process. This issue is fixed in version 2.166.0. |
| MOOS-IvP uFldShoreBroker through 24.8.1 fails to limit the number of claimed communities stored in parallel vectors within ShoreBroker::handleMailNodePing(). A single publisher can supply unbounded distinct community names to grow retained state and per-pass work without limit, causing memory exhaustion and performance degradation. |
| MCP Kotlin SDK is the Kotlin Multiplatform software development kit for the Model Context Protocol. In versions 0.7.0 through 0.12.0, `ReadBuffer.append` in `kotlin-sdk-core/src/commonMain/kotlin/io/modelcontextprotocol/kotlin/sdk/shared/ReadBuffer.kt` writes every chunk of bytes received from the stdio transport into a `kotlinx.io.Buffer` with no size cap. Frames are extracted from that buffer only when a `\n` (0x0a) byte is observed. A peer that streams bytes without ever sending a newline causes the internal buffer to grow indefinitely until the JVM (or the surrounding host process) is OOM-killed. The leak is amplified by `StdioServerTransport` and `StdioClientTransport`, which both queue raw chunks through a `kotlinx.coroutines.channels.Channel<ByteArray>(Channel.UNLIMITED)` and then call `readBuffer.append(chunk)` without backpressure or size guard. This is a remote-pre-auth denial of service whenever an SDK stdio server's stdin is fed by an untrusted or attacker-controlled producer (for example: a host program that exec's the MCP server as a subprocess and pipes through bytes received from a network peer, or a sidecar wrapper that proxies bytes from an HTTP endpoint to the stdio transport). Version 0.13.0 fixes the issue. |
