CVE Watch409.857 no arquivo completo

Vulnerabilidades exploráveis hoje

17,690na visualização atual

Score único que combina CVSS, presença no KEV e EPSS. Cada CVE com sua própria ficha — timeline da publicação até a exploração ativa.

No catálogo KEV1,769

Novos KEV · 24H0

Exploit Today ≥ 701,700

Distribuição · última janela

Crítica
5,527
Alta
21,044
Média
20,197
Baixa
2,086

Filtros

Janela

Severidade

Filtros

Vulnerabilidades1,761–1,800 · 17,690

CVECVSSEPSSKEVRExploitTítuloMod.

CVE-2026-1220387.3 ALT

17.0%

——5A vulnerability was identified in krupalshah EmployeeRecruitmentSystem up to 922ff6d208278282e7c8e36d70d9ae356adec9fe. Affected by this issue is the function mysql_query of the file source/reqformprocessing.php of the component Requirement Board Update Handler. Such manipulation leads to sql injection. The attack can be launched remotely. The exploit is publicly available and might be used. This product does not use versioning. This is why information about affected and unaffected releases are unavailable. The vendor was contacted early about this disclosure but did not respond in any way. This vulnerability only affects products that are no longer supported by the maintainer.3d CVE-2026-1220284.3 MÉD

32.2%

——10A vulnerability has been found in NST Labs Tech Nstbrowser 1.21.0. Affected by this vulnerability is an unknown functionality of the file /agent/info of the component Local Agent API. Such manipulation leads to information disclosure. It is possible to launch the attack remotely. The exploit has been disclosed to the public and may be used. The vendor was contacted early about this disclosure but did not respond in any way.3d CVE-2026-1157709.6 CRÍ

5.9%

——2Pocket ID is an OIDC provider that allows users to authenticate with their passkeys to services. From 2.2.0 until 2.12.0, SendEmailVerification uses the token model in backend/internal/model/email_verification_token.go to bind an email-verification token only to the user's identifier, while UpdateUser can change the account email without invalidating the outstanding token and VerifyEmail marks the current address as verified. An ordinary user can request a token for an address they control, change the account to an arbitrary unused address, and redeem the old token so Pocket ID publishes the new address with email_verified set to true. The default affected configuration uses emailsVerified=false and allowOwnAccountEdit=true; the target address must not already belong to another Pocket ID user. A relying application that trusts a verified email for automatic account matching or provisioning can link the attacker's identity to another user's account, while applications that rely exclusively on the stable OIDC sub claim are not affected by that downstream consequence. This issue is fixed in version 2.12.0.3d CVE-2026-108387—

4.2%

——1radare2 is a UNIX-like reverse engineering framework and command-line toolset. Prior to 6.2.2, attacker-controlled data from several binary format plugins could reach RBinInfo->bclass. Opening or analyzing a crafted Pebble or Sega Mega Drive binary could place a dbg:// URI in bclass, causing radare2 to reopen the input through the debug input/output plugin and ultimately pass attacker-controlled arguments to execvp(). This could execute commands with the privileges of the radare2 process without an explicit debugging command. This issue is fixed in version 6.2.2.3d CVE-2026-102337—

44.1%

——13Webmin is a web-based system administration tool for Unix-like servers. Prior to 2.652, untrusted users with access to the File Manager or Upload and Download modules can direct Webmin's server-side URL download functionality to non-public network addresses because download destinations are not restricted by resolved IP address. This allows requests to unauthenticated intranet services or cloud-provider instance metadata endpoints that are inaccessible from the user's own system. This issue is fixed in 2.652.3d CVE-2026-976939.9 CRÍ

57.0%

——17Yuxi is a large-model-based intelligent knowledge base and knowledge graph agent development platform. From 0.6.0.beta1 until 0.7.1, there is insufficient isolation between execution-capable Docker sandboxes and trusted application services. A low-privileged authenticated user who can execute commands in their own sandbox can reach other tenants' sandbox APIs, access their files and secrets, execute commands in active sandboxes, and enumerate, create, or delete sandboxes through the unauthenticated provisioner without knowing administrator, database, or MinIO credentials. Sandboxes can also reach shared internal services; if shipped default service credentials remain unchanged, an attacker can use them, and if the exposed PostgreSQL account is a superuser, obtain command execution in the PostgreSQL container. These paths can disclose or modify cross-tenant data and sandbox state or cause denial of service through sandbox deletion without a container escape or kernel vulnerability. This vulnerability is fixed in 0.7.1. Existing deployments must separately rotate persisted PostgreSQL, Neo4j, or MinIO credentials that were previously set to documented or shipped defaults because upgrading the Compose files does not automatically rotate them.3d CVE-2026-76893—

21.8%

——7tsParticles is a TypeScript library for creating highly customizable particle animations. Prior to 4.3.2, the HTTP mode of @tsparticles/mcp-server reuses one stateful StreamableHTTPServerTransport for every client. An unauthenticated client that sends the first initialize request without an mcp-session-id claims the shared transport state, causing later independent clients to receive Server already initialized and preventing them from establishing sessions until the process restarts. The first client's session remains functional. This issue is fixed in version 4.3.2.1d CVE-2026-636557.5 ALT

37.6%

——11tsParticles is a TypeScript library for creating highly customizable particle animations. Prior to 4.3.1, the HTTP mode of @tsparticles/mcp-server accepts unauthenticated POST /mcp requests in integrations/mcp-server/src/index.ts and accumulates every request-body chunk in a Buffer[] without a cumulative size limit before Buffer.concat and JSON.parse. A remote attacker can stream an arbitrarily large body to exhaust Node.js process memory and terminate the service, including in the official Docker configuration that exposes HTTP mode on 0.0.0.0:3000. Confidentiality and integrity are not affected. This issue is fixed in version 4.3.1.1d CVE-2026-581286.2 MÉD

23.8%

——7Alan: Health Insurance for Android (com.alanmobile) 1.450.0 contains a path traversal vulnerability that allows a co-resident malicious application to write attacker-controlled files outside the intended directory via the exported ShareReceiverActivity, which uses an unsanitized _display_name value from a shared content URI to construct the destination file path. Attackers can send a crafted ACTION_SEND intent with a _display_name containing directory traversal sequences to overwrite sensitive files within the app's sandbox such as databases, shared_prefs, and WebView data, enabling state tampering or code execution in the app's context without authentication or user interaction.3d CVE-2026-575678.6 ALT

52.0%

——16Readest is a modern, feature-rich ebook reader designed for seamless cross-platform access and an intuitive reading experience. In affected desktop Tauri builds prior to 0.11.10, the apps/readest-app/src-tauri/src/transfer_file.rs commands download_file and upload_file accepted a webview-controlled file_path and passed it to File::create or File::open without checking the Tauri filesystem scope, rejecting absolute paths, or blocking parent-directory traversal. The commands were registered in the global invoke handler for the main, updater, and reader-* windows, so JavaScript executing in the privileged application origin could write attacker-controlled content to any path writable by the process or read and send any path readable by the process. Exploitation requires JavaScript execution in that privileged origin, such as through an application XSS or compromised renderer; whether book-embedded JavaScript can reach the parent origin was not established. The write primitive can overwrite trusted or startup files, while the read primitive can expose keys, tokens, and application secrets. This issue is fixed in version 0.11.10.2d CVE-2026-575669.3 CRÍ

48.0%

——14Readest is a modern, feature-rich ebook reader designed for seamless cross-platform access and an intuitive reading experience. Prior to 0.11.10, the unauthenticated /api/opds/proxy route in apps/readest-app/src/app/api/opds/proxy/route.ts accepted an attacker-controlled URL and fetched it server-side without restricting the scheme, blocking private, loopback, link-local, or metadata hosts, or revalidating destinations during automatic redirect following. The route also accepted caller-controlled outbound headers and Authorization values and reflected the upstream status and response body. A remote attacker could use the service as an open request proxy and, on Node, Docker, or other self-hosted deployments with internal-network access, retrieve cloud metadata or internal-service responses. Cloudflare Workers deployments restrict access to link-local and private destinations, which reduces the internal-network impact there, but the unauthenticated public fetch-and-reflect behavior remains. This issue is fixed in version 0.11.10.2d CVE-2026-575656.5 MÉD

64.1%

——19Readest is a modern, feature-rich ebook reader designed for seamless cross-platform access and an intuitive reading experience. Prior to 0.11.10, the apps/readest-app/src/pages/api/storage/upload.ts route constructed an object key from the authenticated user's identifier and a client-controlled fileName without rejecting path separators or parent-directory traversal. In the R2 backend, apps/readest-app/src/utils/r2.ts placed that key into a new Request URL before signing it, and URL normalization collapsed parent-directory components so the resulting presigned PUT could escape the caller's namespace. An authenticated attacker who knew a target user identifier and book hash could overwrite or poison that user's stored book object and corrupt the associated file_key metadata; the exact cross-tenant escape was verified for the R2 backend. This issue is fixed in version 0.11.10.2d CVE-2026-575626.5 MÉD

37.1%

——11Readest is a modern, feature-rich ebook reader designed for seamless cross-platform access and an intuitive reading experience. Prior to 0.11.10, POST /api/stripe/check in apps/readest-app/src/app/api/stripe/check/route.ts accepted a client-controlled sessionId, retrieved the corresponding Stripe Checkout Session, and granted the paid plan to the authenticated caller without verifying session.metadata.userId. The createOrUpdateSubscription path then updated the plans row keyed by the caller's user identifier rather than the Checkout Session owner. Any authenticated free user with a paid active Checkout Session identifier, including one from the user's own single legitimate purchase, could replay that identifier to upgrade additional accounts without payment. This issue is fixed in version 0.11.10.2d CVE-2026-553688.1 ALT

25.2%

——8h2o is an HTTP server with support for HTTP/1.x, HTTP/2 and HTTP/3. Prior to commit 48b7ae12107fd2b118e4641da49599e9aac8edd8, H2O's access-log formatter could write past its log-line buffer when an affected custom access-log format was configured. The allocation errors affected %{error}x, %q when the query string was empty, and numeric elements whose temporary terminating NUL reached the end of the buffer. A matching request could crash the server, while the effect of the one-byte NUL overwrite beyond a crash was not established. This vulnerability is fixed with commit 48b7ae12107fd2b118e4641da49599e9aac8edd8.3d CVE-2026-553677.5 ALT

27.5%

——8h2o is an HTTP server with support for HTTP/1.x, HTTP/2 and HTTP/3. Prior to commit 4212ce114506a073505fd389db16ad2ba4e582a0, a client could send an incomplete CONNECT-UDP Capsule with an excessively large declared length, causing H2O to keep buffering received bytes without a per-Capsule limit while waiting for the remainder. Deployments that enabled proxy.connect-udp and accepted CONNECT-UDP requests could exhaust server memory and lose availability. This vulnerability is fixed with commit 4212ce114506a073505fd389db16ad2ba4e582a0.2d CVE-2026-543805.4 MÉD

5.8%

——2VitaFit cloud API before 2026-10-30 contained a stored cross-site scripting vulnerability that allows authenticated attackers to upload an SVG image containing malicious JavaScript to the shared storage bucket. Attackers can exploit the application's failure to restrict scripts or force downloads for image/svg+xml content, causing the JavaScript to execute when a victim opens the file via direct browser navigation.3d CVE-2026-54379—

—

——0Rejected reason: This CVE ID has been rejected or withdrawn by its CVE Numbering Authority.10h CVE-2026-543789.6 CRÍ

14.0%

——4VitaFit cloud API before 2026-10-30 contained a broken access control vulnerability that allows any authenticated user to obtain temporary AWS credentials by calling the POST /front/user/s3/token/ endpoint. Attackers can use these improperly scoped credentials to list the shared storage bucket and read or overwrite other users' stored files, enabling unauthorized cross-user data access and manipulation.3d CVE-2026-543776.5 MÉD

8.0%

——2VitaFit cloud API before 2026-10-30 contained an insecure direct object reference vulnerability that allows authenticated attackers to access another user's data by supplying a different user_id or profile_id to endpoints such as GET /front/profile/{profile_id}/report/ and GET /front/profile/{profile_id}/body_index_per_day/. Attackers can leverage predictable user_id and profile_id relationships combined with an account-enumeration weakness to read another account's email address, profile, body-composition report, and measurement history at scale.3d CVE-2026-543768.2 ALT

1.6%

——0VitaFit Android application (com.vt.vitafit) before 2.8.1.0 contains an insecure custom URI scheme vulnerability that allows a malicious application to intercept Fitbit login tokens by registering the same callback scheme (vitafit://vtio.cn). Attackers can capture the 30-day Fitbit authorization token with weight scope delivered in the callback URL to read, create, and delete the victim's weight and body-fat records via the Fitbit Body API.3d CVE-2026-543758.4 ALT

4.3%

——1VitaFit Android application (com.vt.vitafit) before 2.8.1.0 contains an insecure session management vulnerability that allows attackers to extract or intercept unencrypted m_session_id tokens stored in application backups, which remain valid for ten years. Attackers can use a stolen session token to read and modify the victim's account or delete the account's authentication record, preventing the legitimate owner from logging in again.3d CVE-2026-543747.5 ALT

19.0%

——6VitaFit Android application (com.vt.vitafit) before 2.8.1.0 contains a hardcoded credentials vulnerability that allows attackers to extract confidential third-party integration secrets by decompiling the APK. Attackers can retrieve embedded WeChat, Tencent QQ, and Fitbit credentials from the decompiled application to abuse the active third-party application registrations.3d CVE-2026-543733.5 BAI

1.3%

——0VitaFit Android application (com.vt.vitafit) before 2.8.1.0 contains a cleartext traffic vulnerability that allows network-based attackers to intercept or modify unencrypted HTTP communications due to the absence of an Android network security configuration. Attackers on an untrusted network can manipulate support content displayed to the user by tampering with unencrypted HTTP traffic between the app and server.3d CVE-2026-543728.8 ALT

3.5%

——1VitaFit Android application (com.vt.vitafit) before 2.8.1.0 contains an insecure TLS/SSL certificate validation vulnerability that allows network-based attackers to intercept and modify HTTPS traffic by exploiting a custom X509TrustManager that trusts all certificates and a hostname verifier that accepts any hostname. Attackers can perform a man-in-the-middle attack to capture account sessions and health-related API data, and use a captured session to delete the account's authentication record and prevent legitimate login.3d CVE-2026-1222668.7 ALT

19.6%

——6Improper isolation of shared resources in aws-network-policy-agent in Amazon EKS Network Policy Agent before 1.3.7 (VPC CNI before 1.22.3) might allow an authenticated adjacent user to bypass NetworkPolicy ingress enforcement on co-located pods via eBPF conntrack map poisoning using crafted AF_PACKET frames. To remediate this issue, users should upgrade to Amazon VPC CNI Managed Add-on v1.22.3 or later (which includes Network Policy Agent v1.3.7).2d CVE-2026-1220204.3 MÉD

25.0%

——8A security flaw has been discovered in NST Labs Tech Nstbrowser up to 1.2.1. The affected element is the function stopAllBrowsers of the file /api/agent/browser/stopAll of the component Local Agent Management Service. Performing a manipulation results in denial of service. Remote exploitation of the attack is possible. The exploit has been released to the public and may be used for attacks. The vendor was contacted early about this disclosure but did not respond in any way.3d CVE-2026-1220197.3 ALT

21.9%

——7A vulnerability was identified in 6amTech 6valley 16.0. Impacted is the function existingAccountCheck of the file /api/v1/auth/existing-account-check of the component Social Login Flow. Such manipulation of the argument user_response leads to authorization bypass. The attack may be launched remotely. The exploit is publicly available and might be used.3d CVE-2026-1220166.3 MÉD

19.3%

——6A vulnerability was determined in yangzongzhuan RuoYi-Vue up to 3.9.2. This issue affects the function SysPermissionService.getMenuPermission of the file SysPermissionService.java of the component Permission Service. This manipulation of the argument perms causes improper privilege management. The attack may be initiated remotely. The vendor was contacted early about this disclosure but did not respond in any way.3d CVE-2026-1220156.3 MÉD

24.2%

——7A vulnerability was found in NPC-Worldwide npcpy up to 2.1.27. This vulnerability affects the function get_mlx_response of the file npcpy/gen/response.py of the component MLX Provider. The manipulation of the argument model results in improper input validation. The attack can be launched remotely. The exploit has been made public and could be used.3d CVE-2026-1217513.1 BAI

14.1%

——4Node Version Manager (nvm) is a POSIX-compliant shell function for managing multiple node.js versions. In versions 0.39.6 through 0.40.8, `nvm ls-remote` (and other commands that refresh remote LTS aliases) store the version column of the mirror's `index.tab` in `$NVM_DIR/alias/lts/` without validating it, and `nvm_normalize_lts` later evaluates any suffix of an `lts/-N` value inside shell arithmetic `$((N+1))`. Under bash and zsh, arithmetic evaluation expands command substitutions in array subscripts, so a stored value such as `lts/-10*path[$(cmd)1]` runs `cmd` as the user when the alias is later resolved by purely local commands (`nvm ls`, `nvm alias`, `nvm use --lts`, `nvm which lts/`), and in every new shell when the `default` alias is `lts/*`. Exploitation requires the victim to use a hostile mirror -- via a compromised mirror or CDN, a network man-in-the-middle of an `http://` mirror, or a maliciously configured `NVM_NODEJS_ORG_MIRROR` -- and to run an affected command. The same sink is also reachable from a project `.nvmrc`, which is treated as hardening. dash is not affected. Version 0.41.0 only passes an all-digit `lts/-N` offset to arithmetic and drops `index.tab` rows whose version is not a valid version string.3d CVE-2026-1217503.1 BAI

3.8%

——1Node Version Manager (nvm) is a POSIX-compliant shell function for managing multiple node.js versions. In versions 0.40.5 through 0.40.8, `nvm_download_artifact` downloads a release archive directly to its final path in `$NVM_DIR/.cache` and verifies its checksum afterwards; on a checksum mismatch it reports an error but leaves the rejected archive in place. A later `nvm install --offline` of the same version installs that cached file without a checksum, so an archive that was just rejected gets installed and its `node` and `npm` run. This gives code execution to a party who can alter the downloaded tarball (for example a poisoned CDN or cache object) but cannot also forge the `SHASUMS256.txt` beside it, if the victim runs `nvm install --offline` for that version after the visible "Checksums do not match" failure and before any online retry. Version 0.41.0 only puts an archive in the cache once its checksum has passed.3d CVE-2026-1217493.3 BAI

5.5%

——2Node Version Manager (nvm) is a POSIX-compliant shell function for managing multiple node.js versions. In versions 0.5.0 through 0.40.8, the requested version pattern is passed to `grep` as a bare argument, without `-e`, in `nvm_ls_remote_index_tab`, `nvm_remote_version`, and `nvm_ls_cached`. A `.nvmrc` whose first line starts with `-`, such as `-Rohenpm_[A-Za-z0-9]*`, is parsed as grep options; with GNU grep, `-R` searches the current directory recursively (following symlinks) instead of the version list. On a bare `nvm install`, nvm then treats a matched string from those files as the version, prints it, and requests it from the configured mirror inside download URLs, so for example a repository containing a symlink to `../../.npmrc` can leak the user's npm token to the terminal and to the mirror's and any proxy's logs. The disclosure is read-only and limited to the characters `0-9A-Za-z._+-`. Version 0.41.0 passes every version pattern to grep with `-e`.3d CVE-2026-1217483.6 BAI

2.4%

——1Node Version Manager (nvm) is a POSIX-compliant shell function for managing multiple node.js versions. In versions 0.27.0 through 0.40.8, when a project's `.npmrc` sets `prefix`, `nvm use` tells the user to run `nvm use --delete-prefix`, which runs `npm config delete prefix` and `npm config delete globalconfig` in the current directory without naming a config file. npm loads the project `.npmrc` before deciding which file to edit, so that file chooses the target: `userconfig=~/.bashrc` makes npm rewrite `~/.bashrc` as ini (corrupting it), and `location=global` with `globalconfig=` leaves the chosen file at mode 0666, which on a shared machine lets another local user write to it. The attacker (the author of a cloned repository or archive, or a local user who can write a shared parent directory) chooses the file but not the bytes written, and no project code runs. Version 0.41.0 no longer lets a project `.npmrc` select which config file `--delete-prefix` edits.3d CVE-2026-1217475.5 MÉD

7.7%

——2Node Version Manager (nvm) is a POSIX-compliant shell function for managing multiple node.js versions. In versions 0.2.0 through 0.40.8, nvm does not check whether `.nvmrc` is a symbolic link. A repository or archive that ships `.nvmrc` as a symlink to another file (for example `../../.npmrc`, or `/proc/self/environ` on Linux) makes a plain `nvm use`, `nvm install`, `nvm which`, `nvm exec`, `nvm run`, or `nvm-exec` read that file and print its contents in the invalid-`.nvmrc` message (even with `--silent`) or, for a one-line file, in the `Found ... with version <...>` message; with the README's zsh `load-nvmrc` or bash `cdnvm` hook installed, entering the directory is enough. nvm sends the content nowhere; the attacker learns it only where they can see the victim's output, such as a CI log for a contributed branch. Version 0.41.0 still follows a symlinked `.nvmrc`, so deliberate links to another version file keep working, but never prints its content, and accepts it only if it contains a version or the name of an existing alias; the README's bash `cdnvm` hook, which read `.nvmrc` itself, now gets the version from nvm, so users of that hook must also update their copy of it.3d CVE-2026-1217465.5 MÉD

6.3%

——2Node Version Manager (nvm) is a POSIX-compliant shell function for managing multiple node.js versions. In versions 0.40.0 through 0.40.8, `nvm use --save` and `nvm install --save` (or `-w`) write the resolved version to `$PWD/.nvmrc` with `tee`, which follows a symbolic link. A repository that commits `.nvmrc` as a symlink to a path outside itself (git recreates the link on clone) therefore makes `--save` truncate any file the user can write, such as a shell startup file or `~/.ssh/authorized_keys`, and replace its contents with the version string, or create the target of a dangling link; the link itself is left unchanged, so `git status` stays clean. The attacker does not control the content written, so this is data loss, not code execution. Version 0.41.0 refuses, with a warning, to write when `.nvmrc` in the current directory is a symlink, dangling or not, rather than writing through it or replacing it.3d CVE-2026-1217452.5 BAI

2.0%

——1Node Version Manager (nvm) is a POSIX-compliant shell function for managing multiple node.js versions. In versions 0.40.0 through 0.40.8, when `NVM_AUTH_HEADER` is set, `nvm_download` passes the credential to curl or wget as a `--header "Authorization: ..."` command-line argument. Process arguments are visible to every local user (on Linux via `ps` or `/proc//cmdline` unless `/proc` is mounted with `hidepid=`; on macOS via `ps`), so another unprivileged account on the same machine can read the token while nvm is downloading during `nvm ls-remote`, `nvm install`, and similar commands, and gain whatever access it grants on the private mirror. Only users who set `NVM_AUTH_HEADER` on a host shared with other local accounts are affected. Version 0.41.0 passes the header to curl on stdin and to GNU Wget in a private temporary config file. BusyBox wget reads no config file, so when it is the only downloader available the header is still passed as an argument.3d CVE-2026-1154736.5 MÉD

18.3%

——5Mongoose is an embedded web server and network library. Prior to 7.22, mg_path_is_sane in src/util.c read path.buf[1] without first requiring path.len to be at least two and read s[2] when only two bytes remained. Unauthenticated remote requests reaching mg_http_upload or mg_http_serve_dir with the affected short path shapes could therefore trigger one-byte heap out-of-bounds reads. On bare-metal or RTOS targets without memory protection, the adjacent byte could be exposed or influence whether a path was accepted, and instrumented builds could crash. This issue is fixed in version 7.22.2d CVE-2026-1030028.8 ALT

30.0%

——9GPUStack is an open-source GPU cluster manager for AI model serving and GPU instance provisioning. From 2.2.0 until 2.2.2, cluster_apiserver_proxy forwarded an authenticated non-administrator's requests to a selected Kubernetes cluster without a resource-level authorization check. The proxy used a reachable worker's Kubernetes ServiceAccount credentials, so the caller could read sensitive resources, create or modify workloads, or delete resources according to that ServiceAccount's permissions. Exploitation required a Kubernetes-provider cluster with a reachable worker, and attacker-controlled workloads could result in code execution in the target cluster. This issue is fixed in version 2.2.2.2d CVE-2026-1028959.8 CRÍ

26.4%

——8GPUStack is an open-source GPU cluster manager for AI model serving and GPU instance provisioning. Prior to 2.2.0, get_current_user in gpustack/api/auth.py treated a loopback request source as the built-in administrator when force_auth_localhost was false, which was the default. A remote request could appear to originate from 127.0.0.1 because client_ip_getter accepted a same-host reverse proxy peer or a client-supplied X-GPUStack-Real-IP loopback value. The bypass allowed unauthenticated administrative access, including user and model changes, persistent administrator API-key creation, disclosure of worker and cloud-provider secrets, and code execution through POST /v2/inference-backends/from-yaml. Exploitation was observed in the wild against a 2.1.1 deployment. This issue is fixed in version 2.2.0.2d CVE-2026-1028949.1 CRÍ

17.6%

——5GPUStack is an open-source GPU cluster manager for AI model serving and GPU instance provisioning. From 0.7.1 until 2.2.2, the /auth/saml/callback endpoint decoded SAML XML, used its NameID and attributes to resolve or provision a user, and minted a gpustack_session JWT without calling OneLogin_Saml2_Auth.process_response to verify the signature, issuer, audience, destination, or time conditions. Any network client could submit an identity for an existing user, including an administrator, or provision a new non-administrator account, even when SAML was not configured. Response and Assertion identifiers were not consumed, so captured signed responses could also be replayed. This issue is fixed in version 2.2.2.2d