CVE-2026-81176
Svelte devalue is a JavaScript library that serializes values into strings when JSON.stringify isn't sufficient for the job. Prior to 5.9.2,
CVSS
5.3
Medium
EPSS
0.5%
p41
KEV
—
Exploit Today
12
0-100
Published: Sep 16, 2026 · Last modified: Sep 16, 2026 · CWE-770
Not enough EPSS history yet.
Svelte devalue is a JavaScript library that serializes values into strings when JSON.stringify isn't sufficient for the job. Prior to 5.9.2, devalue.parse does not reject out-of-bounds indices that are greater than or equal to values.length in src/parse.js. A specially crafted untrusted payload can make the parser alternate between array representations, producing quadratic work as the payload grows and causing denial of service in applications that parse untrusted devalue data. This issue is fixed in version 5.9.2.
CVECVSSEPSSKEVRExploitTitleMod.
CVE-2026-929157.3 HIG—
——0WWBN AVideo through commit e01e41ecc (no patched version available) contains a broken access control flaw in objects/userVerifyEmail.php. The script disables the login requirement ($global['ignoreUserMustBeLoggedIn'] = 1), takes users_id directly from the query string, and calls User::sendVerificationLink() with no session requirement, no CSRF/global token, no relationship check between caller and target, and no enforceRateLimit() call. The only intended throttle is keyed to the caller's own session, so cookie-less requests are never limited. An unauthenticated remote attacker can therefore cause an arbitrary number of verification emails to be sent to any account ID, and can enumerate accounts and their verification status from the three distinct JSON responses ("Verification Sent", "Already verified", "Unknown error"). In addition, createVerificationCode() invokes $user->setRecoverPass() and saves the user, so each anonymous request writes a live password-recovery token onto the targeted account; that token is embedded in base64 in the verification link emailed to the account owner and is accepted by objects/userRecoverPassSave.json.php as the credential for setting a new password.23hCVE-2026-252817.4 HIG5.2%
——2Transient DOS when processing large or numerous request buffers without sufficient memory allocation validation.1dCVE-2026-823997.5 HIG47.6%
——14CoreDNS is a DNS server written in Go. Prior to 1.14.7, the DNS-over-HTTPS, DNS-over-HTTP/3, DNS-over-QUIC, and DNS-over-gRPC request paths in plugin/pkg/doh/doh.go, core/dnsserver/server_quic.go, and core/dnsserver/server_grpc.go call dns.Msg.Unpack on attacker-controlled DNS section counts before dns.DefaultMsgAcceptFunc validates the fixed header. An unauthenticated client can use DNS name compression and excessive section counts to amplify allocation before the plugin chain, so plugin-level rate limiting cannot prevent concurrent requests from exhausting memory and terminating CoreDNS. The ordinary UDP and TCP listeners are not affected because they validate the header first. This issue is fixed in version 1.14.7.2dCVE-2026-75516—43.5%
——13The RabbitMQ Java client library allows Java and JVM-based applications to connect to and interact with RabbitMQ nodes. Prior to 5.34.0, AMQConnection.start() applies Math.min(maxInboundMessageBodySize, frameMax) after Connection.Tune negotiation even though AMQP defines frameMax value zero as unlimited and ConnectionFactory.DEFAULT_FRAME_MAX is zero. When the client default and server-negotiated value are both zero, the result is passed to Utils.framePayloadLimit(int), which interprets zero as Integer.MAX_VALUE and disables the configured maxInboundMessageBodySize cap. A malicious AMQP server, or a man-in-the-middle attacker able to modify Connection.Tune and inject frames into the connection, can then send an oversized frame of any frame type, causing Frame.readFrom() to allocate a large byte array before content-level validation and potentially terminate the client process through memory exhaustion. This issue is fixed in version 5.34.0.2dCVE-2026-691476.5 MED39.0%
——12vLLM is an inference and serving engine for large language models. Prior to 0.28.0, request bodies for Chat Completions and Responses can set media_io_kwargs.video.video_backend to pynvvideocodec, and MediaConnector.fetch_video forwards that choice to VideoMediaIO even when startup configuration selected a software decoder. The engine's _reserve_mm_ipc_gpu_memory logic budgets decoder memory only from static configuration, so the request-selected VIDEO_LOADER_REGISTRY backend can create a CUDA context, decoder surfaces, and decoded-frame allocations that were not removed from the engine's KV-cache budget. An attacker able to submit video requests to a video-capable GPU deployment with PyNvVideoCodec installed can exhaust shared GPU memory, causing request failures, worker crashes, or denial of service. The first release containing the fix is version 0.28.0.2dCVE-2026-571736.5 MED49.9%
——15vLLM is an inference and serving engine for large language models. Prior to 0.24.0, the input_audio handling path for /v1/chat/completions calls AudioMediaIO.load_bytes or AudioMediaIO.load_file without passing VLLM_MAX_AUDIO_DECODE_DURATION_S to the shared audio decoder. An unauthenticated client can therefore submit a small compressed audio input that expands into a very large float32 PCM allocation, bypassing the duration guard already used by /v1/audio/transcriptions and causing an out-of-memory worker crash. Inline data URLs reach this path without being bounded by VLLM_AUDIO_FETCH_TIMEOUT. The issue affects deployments serving an audio-capable model, and authentication changes only the deployment-specific reachability. This issue is fixed in version 0.24.0.2d