🚨 CVE-2026-101915
@grpc/grpc-js implements the core functionality of gRPC purely in JavaScript, without a C++ addon. Prior to 1.13.6 and 1.14.5, when an application method handler throws an uncaught error, the server includes its error message in the status message sent to the client. The thrown error message is transmitted to the client, causing sensitive information disclosure when the message contains sensitive data. This issue is fixed in versions 1.13.6 and 1.14.5.

🎖@cveNotify

🚨 CVE-2026-102004
Wind River VxWorks 7 prior to 26.09, specific system call arguments can result in memory corruption within the memory management subsystem. Fixed in Version 26.09

🎖@cveNotify

🚨 CVE-2026-86950
An out-of-bounds write issue was addressed with improved bounds checking. This issue is fixed in iOS 26.7.1 and iPadOS 26.7.1, macOS Sequoia 15.8.1, macOS Tahoe 26.7.1. Processing a maliciously crafted file may lead to arbitrary code execution. Apple is aware of a report that this issue may have been exploited in an extremely sophisticated attack against specific targeted individuals on versions of iOS before iOS 27.

🎖@cveNotify

🚨 CVE-2026-87741
The ConvertPlus plugin for WordPress is vulnerable to Deserialization of Untrusted Data in all versions up to, and including, 3.6.3 via the style parameter of the cp_display_preview_modal AJAX action. The vulnerability exists because the action's nonce guard is gated behind an isset() check and fails open when the cp_admin_page_nonce parameter is omitted entirely, no capability check is performed on the callback, and sanitize_text_field() — applied to the $style value before it is concatenated directly into a shortcode string evaluated by do_shortcode() — does not strip shortcode delimiters, allowing an attacker to inject a second, fully attacker-controlled [smile_modal] invocation that causes smile_modal_popup() to pass attacker-supplied base64-decoded bytes to maybe_unserialize() with no allowed_classes restriction. This makes it possible for authenticated attackers, with Subscriber-level access and above, to inject a PHP object. No known POP chain is present in the vulnerable software, which means this vulnerability has no impact unless another plugin or theme containing a POP chain is installed on the site. If a POP chain is present via an additional plugin or theme installed on the target system, it may allow the attacker to perform actions like delete arbitrary files, retrieve sensitive data, or execute code depending on the POP chain present.

🎖@cveNotify

🚨 CVE-2026-93355
LiteLLM contains a weak authentication vulnerability that allows an attacker holding a valid JWT from the configured identity provider to authenticate as any existing user by exploiting an email-based fallback lookup in the JWT authentication flow without verifying the email_verified claim. Attackers can present a token with an unverified email address matching a victim's account to inherit the victim's role, including proxy_admin privileges, and permanently overwrite the victim's stored identity binding to retain persistent unauthorized access to administrative endpoints exposing API keys and user management.

🎖@cveNotify

🚨 CVE-2026-6721
IBM Concert 1.0.0 through 3.0.0 allows an unauthenticated remote attacker can supply specially crafted input that is incorporated into OS commands, resulting in arbitrary command execution on the underlying system. Successful exploitation allows remote code execution with the privileges of the affected application.

🎖@cveNotify

🚨 CVE-2026-6730
IBM Concert 1.0.0 through 3.0.0 is vulnerable to a buffer overflow, caused by improper bounds checking. A local user could overflow the buffer and execute arbitrary code on the system.

🎖@cveNotify

🚨 CVE-2026-6794
IBM Concert 1.0.0 through 3.0.0 has a double free vulnerability that exists due to incorrect memory management. A local attacker can exploit this flaw to corrupt heap memory and execute arbitrary code in the context of the affected process.

🎖@cveNotify

🚨 CVE-2026-6925
IBM Concert 1.0.0 through 3.0.0 could allow a remote attacker to traverse directories on the system. An attacker could send a specially crafted URL request containing "dot dot " sequences ( /.. /) to view arbitrary files on the system.

🎖@cveNotify

🚨 CVE-2026-6544
IBM Concert 1.0.0 through 3.0.0 allows recursive copying of directories without proper controls which can lead to unintentional inclusion of sensitive or unnecessary files and increased attack surface.

🎖@cveNotify

🚨 CVE-2026-100503
Ghidra versions through 12.1.4 contain a heap use-after-free vulnerability in the decompiler's Funcdata::opInsertAfter function caused by stale INDIRECT effect-op references. Attackers can craft a malicious binary with a specific x86-64 sequence that triggers the vulnerability during decompilation, causing the decompile helper process to crash and denying service to analysts and automated analysis pipelines.

🎖@cveNotify

🚨 CVE-2026-100505
Ghidra versions 11.2 through 12.1.4 contain a heap out-of-bounds read vulnerability in StringManager::getCodepoint when decoding multi-byte UTF-8, UTF-16, or UTF-32 characters without validating remaining buffer length. Attackers can craft malicious binaries with constant byte stores ending in multi-byte lead units to trigger out-of-bounds reads that crash the decompiler or leak adjacent heap memory into decompiled output.

🎖@cveNotify

🚨 CVE-2026-100649
vLLM before 0.29.0 contains a resource-limit bypass vulnerability in PyNvVideoCodec decoder allocation where sampler subclass shadowing allows independent counter increments. Unauthenticated attackers can select different sampler subclasses in video requests to exceed configured decoder limits and exhaust unaccounted GPU memory.

🎖@cveNotify

🚨 CVE-2026-100650
vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.

🎖@cveNotify

🚨 CVE-2026-100653
vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for the FunAudioChat and Tarsier2 architectures: the WhisperFeatureExtractor and speech_tokenizer PreTrainedTokenizerFast loads in vllm/model_executor/models/funaudiochat.py and the Qwen2VLConfig.from_pretrained call used by Tarsier2ProcessingInfo in vllm/model_executor/models/qwen2_vl.py. As a result, deployments pinned to a reviewed revision still resolve these behavior-affecting processor, tokenizer, and config artifacts from the repository's default revision, so a later change to the upstream default branch can alter audio preprocessing, speech tokenizer behavior, or Tarsier2 configuration without any change to the operator's configured pin. This is a supply-chain integrity and reproducibility failure for pinned deployments; it is residual to the earlier fix tracked as GHSA-3ww4-5jv9-j5gm / CVE-2026-47155 and does not constitute remote code execution or a trust_remote_code=False bypass. The issue is fixed in version 0.28.0.

🎖@cveNotify

🚨 CVE-2026-100654
vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).

🎖@cveNotify

🚨 CVE-2026-94418
Under WOLFSSL_SMALL_CERT_VERIFY, ProcessPeerCertParse() runs the certificate signature check separately from the parse to keep peak memory down, then merges the two results, but it merged the signature result back only when the parse returned 0, so any parse error hid it. ParseCertRelative() reaches its validity-date, name-constraint and critical-extension checks only after ConfirmSignature() has passed, so splitting the signature check out inverts the precedence that makes "override date errors" a sound policy, and ASN_SIG_CONFIRM_E is never surfaced anywhere. The attacker needs no key material from the real PKI and no CA compromise: a self-made certificate carrying the expected subject name, the trusted CA's subject as its issuer, arbitrary bytes where the signature goes, a validity window in the past and the attacker's own key pair is sufficient. Affected builds define WOLFSSL_SMALL_CERT_VERIFY, which is off by default, is not set implicitly by any platform or preset header, and is not reachable from any CMake option; the autotools routes are --enable-lowresource, --enable-leantls, --enable-tinytls13=cert and --enable-tinytls13=mutualauth, and examples/configs/user_settings_embedded.h reaches it through WC_CFG_SMALL_CERT_VERIFY, which ships as 0, while neither --enable-all nor --enable-distro enables it at all. The application must additionally install a verify callback through wolfSSL_CTX_set_verify() or wolfSSL_set_verify() with WOLFSSL_VERIFY_PEER that returns 1 for ASN_BEFORE_DATE_E or ASN_AFTER_DATE_E; wolfSSL ships this exact shape as myVerify() in wolfssl/test.h under VERIFY_OVERRIDE_DATE_ERR, which examples/client -D selects. An application with no callback, or whose callback returns preverify for date errors, still fails the handshake, and wolfSSL_CertManagerVerifyBuffer() and wc_CheckCertSignature() report ASN_SIG_CONFIRM_E correctly in the same binary. TLS 1.2 and TLS 1.3 are affected in both directions, and DTLS reaches the same function; where the forged certificate is a chain certificate the callback's consent causes it to be cached in the WOLFSSL_CTX certificate manager, so an exposed deployment must restart the context or the process rather than merely reconnect.

🎖@cveNotify

🚨 CVE-2026-94419
Without NO_SESSION_CACHE_REF, wolfSSL_get_session() does not return a session object but a ClientSession reference of the form {row, index, hash(sessionID)} into the process-global SessionCache, and ClientSessionToSession() validates it against that hash alone. Because the TLS 1.2 session ID is chosen by the server and sent in clear, AddSessionToCache() matches any other server's session on the same ID and overwrites the client-side entry with that server's master secret, cipher suite and version, while the handle continues to resolve; nothing on the write path compares the peer, the application's server ID or the WOLFSSL_CTX. Resuming through the handle then produces an abbreviated handshake in which no Certificate message is sent, so neither chain verification nor wolfSSL_check_domain_name() runs, and the attacker is accepted as the original server for the whole of that connection. Affected builds are those leaving NO_SESSION_CACHE_REF, NO_SESSION_CACHE, NO_CLIENT_CACHE and TITAN_SESSION_CACHE all undefined, which includes a plain ./configure, --enable-opensslextra and --enable-opensslall; fifteen integration options define NO_SESSION_CACHE_REF and are therefore not affected, among them --enable-all, --enable-distro, --enable-curl, --enable-nginx, --enable-haproxy, --enable-stunnel, --enable-wpas and the rest of the OPENSSL_COMPATIBLE_DEFAULTS family, and --enable-leanpsk, --enable-leantls, --enable-lowresource and --enable-tinytls13 disable the cache outright. The application must use the legacy reference flow, wolfSSL_get_session() or SSL_get_session() followed by wolfSSL_set_session(); wolfSSL_get1_session() returns the session object itself and is not affected, nor are wolfSSL_SetServerID() lookups. Only TLS 1.2 and below and DTLS 1.2 and below are reachable, since TLS 1.3 and ticket resumption with an empty ServerHello session ID both use a client-chosen cache key. The poisoned entry lives in the process-global cache, so it crosses WOLFSSL_CTX boundaries and persists until the entry is evicted or the session times out, 500 seconds by default. Releases v5.3.0 through v5.9.2 are affected; the fix adds a per-write generation counter to the cache and raises WOLFSSL_CACHE_VERSION from 2 to 3, so a cache persisted by an older build is rejected by a fixed one.

🎖@cveNotify

🚨 CVE-2026-15442
In all builds that make use of (D)TLS, including default builds, there is a series of conditional states during the TLS shutdown which could lead to a heap-use-after free. If an application ended up getting a partial wolfSSL_read() which is sometimes caused by a small user buffer passed in, then called wolfSSL_shutdown for a bidirectional close and attempted to wolfSSL_read() again while the peer continues trying to send data during the shutdown it would lead to a state where a potential heap-use-after free happened.

🎖@cveNotify

🚨 CVE-2026-89102
In wolfSSL versions 5.7.2 through 5.9.2 there is a client-side implementation flaw in RFC 6961, multiple OCSP response stapling, which can lead to certificate forgery. When a wolfSSL client enables OCSP stapling with the HAVE_CERTIFICATE_STATUS_REQUEST_V2 feature and calls wolfSSL_UseOCSPStaplingV2(ssl, WOLFSSL_CSR2_OCSP_MULTI, options), the client accepts any certificate in the peer's chain as a certificate authority without verifying that the certificate is actually authorized to act as one. This means that an attacker who possesses any certificate that chains to a CA trusted by the client (along with its private key) can forge certificates for arbitrary identities that will be accepted as valid by the client. The end entity certificate of the server is stored in the persistent trust store, affecting subsequent connections that reuse the context even when OCSP multi usage is not employed. Found by internal wolfSSL testing.

🎖@cveNotify

🚨 CVE-2026-89133
wolfSSL versions 5.9.2 and earlier contain a flaw in the X.509 certificate validation logic where it fails to properly enforce NameConstraints extensions when there is an unconstrained CA tier between a name-constrained intermediate CA and the leaf certificate. wolfSSL incorrectly accepted certificates for hostnames they shouldn't be allowed to cover, due to a chain-walking state-machine bug that resets the validation state when encountering an intermediate without NameConstraints, thereby bypassing cryptographic delegation controls. This defect exists in the default build configuration that makes use of certificates where name constraint extensions are used. Thanks to Jack Lloyd, PathDiff, and Ben Smyth for reporting the issue.

🎖@cveNotify