CVE-2026-80322 is a race condition vulnerability in Ping Identity's PingAuthorize component operating in Gateway mode. Under certain conditions, an asynchronous upstream request may continue after the 30-second gateway timeout, causing a late response to be incorrectly sent to a subsequent request. This can lead to unauthorized information disclosure between separate requests. The issue affects multiple versions prior to fixed releases and has a high severity rating.
@fastify/multipart is the official Fastify plugin for handling multipart/form-data uploads. In versions 3.0.0 through 10.1.3, temporary files written by the saveRequestFiles helper are removed only by a response hook that runs when the response finishes or errors. When a client completes an upload and then disconnects before the handler writes the reply, for example while the handler scans, hashes or forwards the file, the response only emits a close event, the cleanup hook never runs, and the saved files are left on disk. An unauthenticated client can repeat this cheaply to exhaust disk space and deny service to applications that save uploaded files to disk. The issue is fixed in @fastify/multipart 10.1.4, and users should upgrade to 10.1.4 or later. As a workaround, avoid saving request files for untrusted clients, or clean the saved files from a close handler on the raw response so removal also happens when the client disconnects.
Missing Release of Resource after Effective Lifetime vulnerability in elixir-grpc grpc_server allows a client to stall all later calls on a shared HTTP/2 connection via calls that the server refuses while their DATA frames are in flight.
The stream handler grpc_stream_h (grpc_server/src/grpc_stream_h.erl) requests flow from Cowboy only for request data that has arrived. When the server refuses a call early, for example because the body exceeds max_body_size or the method is not implemented, Cowboy drops the DATA frames still in flight for that stream. Their bytes are not returned to the connection flow-control window. After enough refused calls the window is empty. No call on that connection can then send a request body, and each call waits until its deadline. Other connections are not affected.
This issue affects grpc_server: from 1.0.0 before 1.1.0.
P-Net before 1.0.4 keeps the RPC session that is allocated for a fragment of a request with an unknown activity UUID, without a timeout, when the rest of the request never arrives. A session that does not get an Application Relationship is also kept after the response. Repeated fragments or requests with different activity UUIDs exhaust the available sessions, one UDP packet per session. Once they are exhausted the device no longer answers a Connect request until it is restarted. An unauthenticated attacker with ADJACENT network access can use this to cause a denial of service.
Active Storage makes it simple to upload and reference files in cloud services and attach those files to Active Record objects. From 8.0 until 8.0.5.2 and 8.1.4.1, when Active Storage serves a file through its proxy controllers, ActiveStorage::Streaming#send_blob_stream handles an error from the storage service by calling head to send a 404 or 500 response. If the error occurs after the first chunk has been written, the response is already committed, and head replaces the response body while the server is still reading the original ActionController::Live buffer. That buffer is never closed, so the server thread handling the request blocks until the process restarts. Repeated failed downloads could exhaust the server's threads, possibly resulting in a DoS vulnerability. Applications are affected when they store files on a service that streams a download in several requests, such as Amazon S3, an S3-compatible service or Google Cloud Storage, and draw Active Storage's default routes. The proxy routes are drawn whether or not proxy mode is the default for the application. A file deleted while it is being served, or any other storage error partway through a download, triggers the condition. This issue is fixed in versions 8.1.4.1 and 8.0.5.2.
On affected platforms running Arista EOS with VPLS-BGP enabled, an attacker who has established a BGP session with the PE router can affect the route advertisements of VPLS pseudowires, causing a churn of the established routes with other legitimate BGP peers.
A flaw exists in the server-side SSPI Kerberos authentication handler on Windows Enterprise builds of MongoDB. When processing certain authentication requests, the server fails to properly release an underlying system resource due to an internal tracking error. An unauthenticated user can trigger this by repeatedly sending authentication requests using the Kerberos mechanism, eventually exhausting available resources and causing all subsequent Kerberos authentication attempts to fail. Server restart is required to recover.
CVE-2026-119636 is a vulnerability in Devolutions Server versions 2025.3.2.0 through 2026.2.22.0 where expired privileged access checkouts in the PAM scheduled check-in task are not properly released. This allows an authenticated user with checkout rights to retain access to a PAM account beyond the intended expiration if there is a backlog of 21 or more newer active checkouts, preventing password rotation and checkout check-in.
Eclipse ThreadX NetX Duo versions 6.0 through 6.5.2.202603a contain a vulnerability in the NetX Secure DTLS server where sessions are allocated from a fixed pool for unrecognized addresses before handshake completion and are never released if the handshake stalls. This allows a remote unauthenticated attacker to exhaust the session pool by sending spoofed datagrams, causing denial of service for legitimate clients until sessions are reset.
CVE-2026-90870 is a vulnerability in Progress Software WS_FTP Server affecting versions prior to 2024.0.7 and 2025.0.4 on Windows. It involves uncontrolled resource consumption due to missing release of backend receiver sessions in the Ad Hoc Transfer component. This can lead to server resource exhaustion and service degradation. The issue has a medium severity with a CVSS score of 5.3 and does not impact confidentiality or integrity, only availability. Fixed versions 2024.0.7 and 2025.0.4 are available.
RT-Labs P-Net versions before 1.0.3 contain a resource management vulnerability where RPC sessions allocated for requests naming unknown interfaces or object UUIDs are not released or timed out. This leads to exhaustion of available sessions, causing the device to stop responding to new RPC requests until restarted. This can be exploited remotely without authentication to cause a denial of service.
RT-Labs P-Net versions before 1.0.3 contain a resource management vulnerability where Application Relationships (AR) and RPC sessions are not released after rejecting RPC requests with incorrect endianness data representation. This leads to session exhaustion, causing the device to stop responding to new RPC requests until restarted. The flaw can be exploited remotely without authentication to cause a denial of service.
CVE-2026-118262 is a resource management vulnerability in the labstack Echo Go web framework affecting versions from 4.15.0 up to but not including 4.16.1, and from 5.0.0 up to but not including 5.4.1. The Proxy middleware can retain handlers, goroutines, and network connections after a WebSocket client disconnects if the upstream connection remains open. This leads to resource accumulation and potential denial of service. The issue is fixed in versions 4.16.1 and 5.4.1.
CVE-2026-95108 is a resource exhaustion vulnerability in Eclipse Foundation's Eclipse ThreadX affecting versions 6.0.1 through 6.5.1.202602a. The flaw arises because the Module Manager's thread-delete dispatcher releases only one of two allocated memory blocks per user-mode thread, causing the kernel stack allocation to remain until the entire module is stopped. This leads to gradual depletion of a finite shared object pool, eventually preventing creation of new kernel objects and user-mode threads system-wide.
CVE-2026-59951 is a medium severity vulnerability in the modelcontextprotocol python-sdk affecting versions 1.8.0 through 1.29.1 and 2.0.0 through 2.1.1. The issue involves uncontrolled resource consumption due to a Streamable HTTP server in stateful mode that never reclaims session state, allowing unlimited sessions to accumulate and consume memory. This can lead to server memory growth without bound until a restart occurs. The vulnerability is fixed in versions 1.30.0 and 2.2.0, which introduce session idle timeouts, maximum session limits, and proper session cleanup.
Libreswan versions 5.3 through 5.3.2 running in FIPS mode have a vulnerability where malformed certificates are rejected but not removed from the NSS cache. This causes legitimate certificates with the same serial number to be rejected, potentially locking out legitimate peers until the service restarts. The issue stems from a missing call to CERT_DestroyCertificate() in the add_decoded_cert() function, related to an incorrect fix for a previous vulnerability (CVE-2026-14957).
CVE-2026-16678 is a medium severity vulnerability in IBM Virtualization Management Interface firmware versions FW1110.00 through FW1110.31, FW1120.00 through FW1120.01, and FW1060.00 through FW1060.81. It involves improper resource management that could allow a remote authenticated attacker to cause a denial of service by exhausting connection resources.
CVE-2026-107834 is a medium severity vulnerability in OWASP Coraza WAF versions 3.0.0 through 3.7.x. It involves uncontrolled resource consumption due to the handling of multipart file uploads, where temporary file descriptors remain open until the entire request completes. An unauthenticated attacker can exploit this by submitting many minimal file parts, exhausting the process file descriptor table and causing failures in creating temporary files, blocking legitimate uploads and preventing the process from opening files or sockets. This issue is fixed in version 3.8.0.
A missing release of resources in the illumos name service cache daemon (nscd) allows a local user to exhaust kernel memory. The nscd door server procedure, switcher() in usr/src/cmd/nscd/nscd_frontend.c, does not close file descriptors that are passed with a door call but not used by the request, and the main nscd door at /var/run/name_service_door accepts passed descriptors from any user in its zone. Because nscd also runs with an unlimited file descriptor limit, an unprivileged local user, including one in a non-global zone, can repeatedly pass a descriptor to its zone's nscd in a door_call() loop, causing the file descriptor table of nscd to grow without bound in kernel memory. This causes a denial of service of nscd and can render processes in all zones on the host unresponsive. The flaw has existed since 2006 (illumos-gate commit cb5caa98), and affects any illumos distribution prior to illumos-gate commit af810a72.
> This issue was posted by Codex Desktop using gpt-6.1-sol on behalf of David.
### Summary
Applications that wrap a model with `ConcurrencyLimitedModel` or `limit_model_concurrency` can permanently lose shared concurrency capacity when a streamed request releases its slot from a different task than the one that acquired it. This can happen when a stream ends early, and also when a stream is fully consumed using the default `stream_text()` debouncing.
In an application that exposes an affected streaming endpoint to network clients and shares a long-lived model limiter across requests, a client can repeatedly start a stream and disconnect. The completed requests retain their slots, eventually preventing subsequent requests that share the limiter from proceeding.
Agent-level `max_concurrency` and non-streaming model requests are not affected by this defect.
### Details
The built-in limiter uses `anyio.CapacityLimiter`, which associates each acquired slot with its borrowing task. Pydantic AI's streaming lifecycle can acquire the slot on the task consuming the stream and run cleanup on another internal task. The limiter rejects that release, so the slot remains occupied even after the request has ended. Cleanup can raise a `RuntimeError`; a later request on the borrowing task can also fail because that task still holds a slot.
Early termination includes stopping iteration, a consumer exception, and cancellation. Fully consuming `stream_text()` with its default `debounce_by=0.1` can also reach the cross-task release path. Fully consumed streams must therefore not be assumed safe.
### Mitigation
Upgrade to a patched release of `pydantic-ai` or `pydantic-ai-slim`. If you cannot upgrade yet, use the agent-level `max_concurrency` setting instead of a concurrency-limited model, or avoid streaming runs through a concurrency-limited model.