The internet is broken right now.
44 critical CVEs without a fix in the last 48 hours. Somewhere a staging server is public. It always is. 46 more already have a fix that someone still has to apply.
-
CVE-2026-126398
vLLM from 0.28.0 before 0.30.0 contains a reachable assertion vulnerability in GPUModelRunner._init_xdrope_positions that allows remote attackers to crash the engine by submitting embeddings-only requests. Attackers can send a /v1/completions request containing only prompt_embeds to XD-RoPE models with --enable-prompt-embeds, terminating the EngineCore and API server for all clients.
-
CVE-2026-126387
n8n before 2.46.1 contains an arbitrary file read vulnerability that allows authenticated members to read server files by supplying absolute paths to the Browser Use browser_upload tool. Attackers with a connected remote browser session can have Playwright read files such as instance configuration from the n8n server and retrieve their contents from the page.
-
CVE-2026-126305
EspoCRM before 9.3.11 and 10.x before 10.0.9 contains a stored cross-site scripting vulnerability in Knowledge Base articles that lets users with Knowledge Base create or edit access attach a JavaScript file and add a crafted link to the article body. When an administrator clicks the link, the script runs in their session and can create an administrator account with attacker-chosen credentials.
-
CVE-2026-126917
Memory corruption in V8 in Google Chrome prior to 157.0.8103.78 allowed a remote attacker to potentially execute arbitrary code inside the sandbox via a crafted HTML page. (Chromium security severity: High)
-
CVE-2026-126250
A incorrect authorization vulnerability in Fortinet FortiAuthenticator 8.0.0 through 8.0.3, FortiAuthenticator 6.6.0 through 6.6.10 may allow an unauthenticated attacker to cause a denial of service via HTTP requests.
-
CVE-2026-126774
@tinacms/mdx before 2.4.2 contains a code injection vulnerability in its MDX serializer, which writes rich-text values into MDX expressions without escaping them. Authenticated content editors can save crafted values that break out of list field or highlightColor expressions in the stored .mdx file, executing JavaScript in visitors' browsers or on build and server-side rendering hosts.
-
CVE-2026-111597
The Forminator Forms WordPress plugin before 1.58.4 does not verify that a Stripe Checkout payment has completed before returning a pending submission's after-payment response, allowing unauthenticated visitors to obtain the post-payment redirect URL and thank-you message of a payment-gated form without paying.
-
CVE-2026-119892
Incorrect authorization in Windows Installer allows an authorized attacker to elevate privileges locally.
-
CVE-2026-11316
A out-of-bounds write vulnerability exists in the dr_wav drwav__read_cue_to_metadata_obj functionality of dr_libs (version(s): 0.14.5). A specially crafted .wav file can lead to an out-of-bounds write. An attacker can provide a malicious file to trigger this vulnerability.
-
CVE-2026-120113
IBM Aspera Connect 1.6.1 through 4.2.19 could allow a remote attacker to obtain sensitive information and modify host trust configuration due to missing authentication and overly permissive cross-origin access controls in the local HTTP service. By persuading a victim to visit a malicious web page, an attacker could perform unauthorized cross-origin requests to read and modify configuration.
-
CVE-2026-123584
The hickory-net crate in Hickory DNS 0.26.0 before 0.26.2 stops listening and returns an error when a client UDP connection receives a datagram that cannot be parsed as a DNS message. Remote attackers who guess the query's source port, without knowing the transaction ID, can spoof malformed responses to abort client queries, causing denial of service.
-
CVE-2026-123327
In JetBrains YouTrack before 2026.2.19562 missing authorization in the saved query endpoint allowed disclosure of issue descriptions
-
CVE-2026-111770
The Model Context Protocol TypeScript SDK provides TypeScript client and server implementations of the Model Context Protocol. From 2.0.0 until 2.3.0 in @modelcontextprotocol/server, applications that shared one stateless HTTP transport or one McpServer or Server instance among multiple clients could deliver responses and notifications to the wrong client. The affected surface is an MCP server reached over HTTP by multiple clients. The trigger is reuse of one stateless transport or server instance. Messages were associated with a connection using only the JSON-RPC request identifier. SDK clients start request identifiers at zero, so collisions occur in normal operation. Cross-client delivery exposes one client's responses and notifications to another client. The issue requires an HTTP endpoint reachable by more than one client and an application that constructs the server or transport once at startup instead of once per request or session. stdio servers and factory-created per-call servers are not affected. Upgrading causes reused instances to fail and therefore requires an application code change. This issue is fixed in @modelcontextprotocol/server 2.3.0.
-
CVE-2026-97174
A flaw was found in the Client-Initiated Backchannel Authentication (CIBA) implementation of Keycloak. The CIBA endpoint fails to properly validate the sub parameter, allowing it to be stored as an additional request parameter. During the token issuance process, this injected value overwrites the legitimate subject identifier. This allows an attacker to authenticate as themselves but receive security tokens belonging to a different victim user, leading to unauthorized account access.
-
CVE-2026-74759
Google::Auth versions 0.12 for Perl allow server side request forgery via unvalidated credential_source URLs in AWS external account credentials. An external_account configuration with the AWS subject token type selects Google::Auth::ExternalAccountCredentials::Aws. Its retrieve_subject_token sends requests to the imdsv2_session_token_url, url and region_url fields of credential_source without the host check that the other credential classes apply. Any caller that builds credentials from a configuration it does not fully control issues GET and PUT requests to hosts the configuration names, including internal services and link-local metadata endpoints.
-
CVE-2026-62162
Hydra is a layer-two scalability solution for Cardano. Prior to 2.1.0, the `Claim` branch in `hydra-plutus/validators/deposit.ak` checks only that `datum.head_id` matches the attacker-controlled `redeemer.currencySymbol` and that the transaction is valid before the deposit deadline. It neither requires spending the Head state-token UTxO in the same transaction nor constrains outputs. After a deposit made through `POST /commit` is confirmed, any third party able to build a Cardano transaction can observe and consume the pending deposit UTxO before its deadline and redirect the assets to an arbitrary address. A standalone deposit spend invokes no Head validator, bypassing its safety checks; historical head-ID binding and the Aiken rewrite retained this missing invariant. The default deadline is three times depositPeriod, so the described attack applies to pending deposits before that deadline. This issue is fixed in version 2.1.0.
-
CVE-2026-104383
Improper Control of Generation of Code in Pigeon MFA service in EagleCloud Aquila versions through 2.6.7.2 on on-premises platforms allows authenticated remote attackers to execute arbitrary code via code injection
-
CVE-2026-77167
Abusing the federated share endpoint, an attacker can write arbitrary rows into the oc_share_external database table. These rows persist indefinitely — there is no automatic cleanup mechanism. Three distinct denial-of-service conditions follow from this write primitive, all triggered from the same unauthenticated source and all self-renewing on a 24-hour cycle for as long as the injected rows remain in the database.
-
CVE-2026-106076
A vulnerability allowing an authenticated organization-transfer peer to inject SQL into the configuration database of the source Veeam Backup for Microsoft 365 server, leading to cross-tenant read and write of configuration data and remote code execution as the PostgreSQL service account.
-
CVE-2026-106070
A flaw in the Node.js `zlib.ZipFile` API can cause it to return bytes that do not belong to the requested entry after the local and central header validation fails. When `zipEntry.modified` is read on an archive whose local header does not match its central directory, either directly or while compacting the archive, the validation error is caught but the parsed local header is left cached without a content offset. A later content read then returns attacker-controlled bytes from the start of the file instead of rejecting the archive, and CRC-32 verification still passes because the attacker controls the central directory CRC. A Node.js-based scanner can therefore approve different content from what another ZIP implementation extracts from the same archive. This vulnerability affects Node.js **26.x**.
-
CVE-2026-79886
An unauthenticated buffer overflow vulnerability exists in the PAPI protocol of AOS-8. Successful exploitation could allow an unauthenticated remote attacker to disrupt the availability of the affected service on the controller, with a limited impact to its integrity.
-
CVE-2026-130647
@fastify/http-proxy is a Fastify plugin that proxies HTTP and WebSocket traffic to an upstream server. In versions from 9.0.0 up to but not including 11.6.10, when the plugin is registered with websocket support enabled, a WebSocket upgrade that is pipelined behind another HTTP request on the same cleartext connection arrives while that request's response is still attached to the socket. The shared upgrade listener sees the socket is already claimed and returns early without taking the upgrade, but it does not attach an error listener to the raw socket, and Node.js has already removed its own error listener before emitting the upgrade event. A later socket error on that connection, such as a client connection reset, is therefore unhandled and terminates the Node.js process. A single unauthenticated connection that pipelines an upgrade behind a normal request and then resets is enough to crash the server, and it can be repeated to keep the process down. Upgrades over TLS are not affected. The issue is fixed in @fastify/http-proxy 11.6.10, and users should upgrade to 11.6.10 or later.
-
CVE-2026-129500
@fastify/http-proxy is a Fastify plugin that proxies HTTP and WebSocket traffic to an upstream server. In versions from 9.0.0 up to but not including 11.6.10, when the plugin is registered with websocket support enabled, it takes every upgraded socket and dispatches the request through the Fastify router before the WebSocket library takes ownership of the socket. Node.js removes its own error listener from the socket before emitting the upgrade event, and the plugin does not attach one of its own during routing, so a socket error such as a client connection reset that arrives before the WebSocket library takes over is unhandled and terminates the Node.js process. The error window is open whenever anything asynchronous runs between dispatch and the route handler, such as an async hook on the proxied route or any unrelated async route in the application. A single unauthenticated upgrade request followed by a connection reset is enough to crash the server, and it can be repeated to keep the process down. The issue is fixed in @fastify/http-proxy 11.6.10, and users should upgrade to 11.6.10 or later.
-
CVE-2026-130157
RepoManager is a self-hosted deb/rpm package repository. Prior to 5.13.3, an attacker who controls an upstream RPM repository mirrored by a RepoManager instance can provide a package filename through the primary.xml <location href> field; the mirror passes that filename into a shell-interpreted RPM query, allowing shell metacharacters to execute commands as the www-data service account. The target must mirror an upstream repository controlled by the attacker, and the default "Check GPG signatures" option triggers the vulnerable call. The command executes before a later signature-verification failure, so a valid signed package is not required. Fixed in 5.13.3.
-
CVE-2026-77380
Bolt is a fully open source, adaptable platform for building and running modern websites. Prior to 6.1.5, the /bolt/async/upload-url and /bolt/async/embed endpoints validate only a submitted URL before fetching it server-side. The fetchers then follow redirects without revalidating each destination, allowing an authenticated user to bypass UrlSafetyChecker and cause the server to request internal or cloud-metadata addresses. This issue is fixed in version 6.1.5.
-
CVE-2026-22309
The visitor tracking of JTL-Shop automatically registers unknown crawler user agents without filtering the value, and the stored value is rendered without output encoding on several pages of the administration backend (search engine statistics, crawler dashboard widget, crawler edit form, sitemap export download list and campaign details). An unauthenticated remote attacker can store script code with a single request. The code is executed in the browser session of an administrator who opens one of these pages, allowing actions to be performed with the administrator's privileges.
-
CVE-2026-129700
In Eclipse Ditto versions 2.4.0 to 3.9.8, Thing fields a requester is not allowed to read can be disclosed through the `condition` and `live-channel-condition` headers, which are evaluated against the full Thing: the READ check of `live-channel-condition` is skipped when a `condition` header is also present on Thing query commands (3.0.0 to 3.9.8); both headers are evaluated without any authorization on WoT Thing Description requests (`Accept: application/td+json`) when WoT TD permission filtering is not active (by default in 3.0.0 to 3.8.14, in 2.4.x with WoT integration enabled, and in 3.9.x with filtering disabled); the `condition` of live messages and live events is evaluated without a READ check, and of claim messages without any permission (3.1.0 to 3.9.8); and the twin response of a Thing query with a non-matching `live-channel-condition` is returned without filtering it by the policy (3.0.0 to 3.9.8). This allows an authenticated user to read Thing fields they are not allowed to read, in some cases without any permission on the Thing.
-
CVE-2026-55505
Wavelog is web-based amateur radio logging software. Prior to 3.2.4, the application/controllers/Lotw.php import() workflow passes the lotw_download_url configuration value to curl without scheme validation, accepts file:// content, and writes the response to a predictable path under ./uploads/ that can be retrieved through an unauthenticated HTTP request. A standard user can trigger the LotW import, although changing lotw_download_url normally requires installer administrator access, and related vulnerabilities can remove that prerequisite. The same missing status-validation pattern affects the LotW lotw_users fetch helper and per-user download helper. This exposes files readable by the www-data user, including environment data, mounted secrets, and application configuration. This issue is fixed in version 3.2.4.
-
CVE-2026-95823
tcp_pkt_linearize() in subsys/net/ip/tcp.c is used by th_get() whenever a received TCP header is not contiguous in a single net_buf. While pulling the header bytes out of the following fragments, the loop called net_buf_unref() on an emptied fragment without first detaching its frags pointer. net_buf_unref() walks the whole fragment list, so the payload buffers still linked behind that fragment were freed as well, and the stale pointer was then reinstalled into the packet as buf->frags, producing a use-after-free and a later double free during packet teardown. The same loop also advanced to second->frags unconditionally: when the fragment still held payload after the header remainder was pulled, it was spliced out of the packet without ever being unreferenced, permanently leaking that buffer and silently dropping the data it carried. The defect is reached from tcp_recv() via th_get() on every received TCP segment, before connection lookup or any header validation, so a single unauthenticated segment that matches a listening socket is sufficient. An attacker only needs the TCP header to straddle a net_buf boundary, which they can arrange from off-link: IPv6 extension headers are accepted on receive and their length is added to the header offset by th_get(), so a Destination Options header of a suitable length places the TCP header across the boundary of the fixed-size receive buffers (CONFIG_NET_BUF_DATA_SIZE). Packet chains assembled from several sources - IPv4/IPv6 reassembly, 6LoWPAN/802.15.4 reassembly, tunnel or virtual interfaces that strip an outer header, or drivers producing scatter-gather chains - can additionally present a short interior fragment that the loop empties exactly, which is the case that frees the remainder of the packet while it is still in use. The reliably reachable consequence is exhaustion of the receive buffer pool: each crafted segment leaks one net_buf from the fixed CONFIG_NET_BUF_RX_COUNT pool, the buffers are never reclaimed, and once the pool is empty the device can no longer receive network traffic until it is rebooted. Where a fragment is emptied exactly, the impact rises to memory corruption: freed pool buffers are written through by the remainder of the loop, their contents may be delivered to the application, and the packet is double-freed on teardown. The fix saves and clears second->frags before the net_buf_unref() so only the emptied header fragment is released, and advances to the next fragment only when second->len == 0, keeping any fragment that still holds payload in the packet. This restores the detach-before-unref ownership pattern used elsewhere in the stack, for example in net_pkt_trim_buffer().
-
CVE-2026-87850
nVision sends user's token via query parameters of GET requests. An attacker who obtains valid token can gain access to the victim's account and perform actions on their behalf. This issue was fixed in version 18.
-
CVE-2026-131857
Supergateway 4.2.0 through 4.2.2 contains a denial of service vulnerability in stdio-to-SSE mode that allows unauthenticated attackers to crash the gateway via prototype-inherited sessionId values. Attackers can send a POST message request with sessionId set to toString or constructor, triggering an unhandled TypeError that terminates every attached MCP session.
-
CVE-2026-130835
Storyden before 1.26.17 contains a missing authorization vulnerability in the AccountRemoveRole handler that allows non-administrator users with MANAGE_ROLES to remove administrator roles. Attackers can send DELETE requests to the account roles endpoint to demote other administrators, potentially leaving the instance with no administrator.
-
CVE-2026-121864
In nanobot, when sandboxed exec is enabled, the shell executable used to run commands can be selected by the caller and is launched before the sandbox is applied. An attacker who can induce the agent to create an executable in the workspace and select it as the shell can cause that executable to run outside the sandbox (e.g. Linux bubblewrap), with the operating-system permissions of the nanobot process. This issue affects nanobot (nanobot-ai) from 0.2.1 before 0.4.1.
-
CVE-2026-110178
IBM ContextForge MCP Gateway 1.0.0 through 1.0.11 is vulnerable to improper access control in the bulk import path. A remote authenticated attacker holding a team-scoped `admin.import` permission can overwrite the endpoint URL, request headers, and authentication credentials of public tools, resources, and prompts owned by other tenants by sending a crafted POST request to `/v1/import` with `conflict_strategy=update`, redirecting subsequent AI gateway invocations to an attacker-controlled host.
-
CVE-2026-63164
Omnispace Agora Project through 26.10.0 contains an insecure direct object reference vulnerability that allows authenticated attackers to access another user's private mail by supplying a crafted reloadMailTypeId parameter to Ctrl::getCurObj() without a readRight() ownership check in CtrlMail.php. Attackers can manipulate the mail identifier value, such as reloadMailTypeId=mail-1, to retrieve the title and body of private messages belonging to other users, resulting in unauthorized disclosure of sensitive communications.
-
CVE-2026-128334
listmonk 6.2.0 before 7.0.0-beta.2 contains a missing authentication vulnerability that allows unauthenticated attackers to submit Azure Event Grid bounce webhooks when no shared secret is configured. Attackers can POST forged EmailDeliveryReportReceived events with Bounced status to /webhooks/service/azure to blocklist, unsubscribe, or delete arbitrary subscribers.
-
CVE-2026-126379
Vvveb 1.0.8.6 contains an incomplete server-side request forgery fix that allows authenticated admin users to bypass DNS pinning by requesting non-443 URLs. Attackers can supply a rebinding hostname to the editor oEmbedProxy url parameter to read loopback services, internal panels, and cloud metadata credentials.
-
CVE-2026-116322
Improper locking in PostgreSQL startup processing allows an unauthenticated network client to trigger a server restart, via arranging for authentication_timeout to arrive while it holds a spinlock during cancel handling. Within major version 18, minor versions before PostgreSQL 18.7 are affected. Versions before PostgreSQL 18 are unaffected.
-
CVE-2026-8333
A stored cross-site scripting vulnerability in Microchip the TimePictra platform could allow an authenticated, low-privileged attacker to store malicious script content that executes when another user views the affected page
-
CVE-2026-0327
A privilege escalation vulnerability in the GlobalProtect gateway of Palo Alto Networks PAN-OS® software enables an authenticated, lower-privileged user to obtain the privileges and configuration of a higher-privileged user. Cloud NGFW and Panorama are not affected by this vulnerability.
-
CVE-2026-126561
Semantic MediaWiki 2.5.0 before 7.4.1 contains a missing authorization vulnerability that allows editors without the edit-protection right to lock pages by adding an Is edit protected annotation. Attackers can add [[Is edit protected::true]] to pages or #set it in widely transcluded templates, so the TitleQuickPermissions hook blocks other editors and administrators from editing, moving or deleting them indefinitely.
-
CVE-2026-120803
A use-after-free issue exists in MongoDB’s SBE (Slot-Based Execution) implementation of a window function accumulator used by certain $setWindowFields aggregation pipelines. Under specific conditions, the accumulator can retain a dangling pointer to memory that has since been freed. When this stale pointer is later dereferenced during serialization of the carried-forward value, it can result in a heap out-of-bounds read that may expose adjacent heap contents and crash the mongod server. An authenticated user with readWrite privileges on a database can trigger this issue by crafting a specific aggregation pipeline.
-
CVE-2026-125111
Rack is a modular Ruby web server interface. Prior to 2.2.25, 3.1.23, and 3.2.8, Rack::Multipart::Parser can retain parsed data in its internal StringScanner buffer when reads end inside or immediately after multipart headers. An unauthenticated attacker can align many small parts with rack.multipart.buffer_size boundaries so the buffer grows with the request despite per-field, header, and preamble limits, causing a worker to exhaust memory. The attack requires only a well-formed multipart body and affects applications that parse multipart requests. This issue is fixed in versions 2.2.25, 3.1.23, and 3.2.8.
-
CVE-2026-117545
IBM webMethods Integration Server 12.1 could allow a remote authenticated attacker to bypass security restrictions due to improper authentication.
-
CVE-2026-124436
OpenRemote before 1.32.2 contains an improper authorization vulnerability that allows realm administrators to access and modify global resources by calling admin endpoints lacking super administrator checks. Attackers can change the global keycloakUrl via /configuration/manager to redirect other realms' console authentication for credential phishing, inject CSS, replace map tiles, and delete logs.
-
CVE-2026-121415
Parse Server before 8.6.111 and 9.0.0 before 9.10.6-alpha.3 on MongoDB skips field type validation for null-valued dot notation keys, allowing sub-documents to be written beneath Pointer, Number, String, Date, or Boolean fields. Any client with write access, including unauthenticated clients where public writes are permitted, can corrupt field values, make objects unreadable via internal date fields, or break atomic operations on number fields.
-
CVE-2025-8574
The Database Cleaner plugin for WordPress is vulnerable to arbitrary table deletion in all versions up to, and including, 1.4.7. This is due to a lack of restriction on database tables that can be deleted. This makes it possible for authenticated attackers, with Administrator-level access and above, to delete core WordPress tables.
-
CVE-2026-77819
IBM Langflow OSS 1.0.0 through 1.12.3 could allow a remote authenticated attacker to overwrite arbitrary files due to improper limitation of a pathname to a restricted directory.
-
CVE-2026-107034
Insecure permissions in LeafPad / Leaf Note v1.21 (versionCode 21) allows unauthorized attackers to arbitrarily read another user's note data.
-
CVE-2026-107038
Insecure permissions in Yet Another UPnP Controller versionCode 50200 allows unauthorized attackers to invoke an exported component, leading to arbitrary data access and control of application data.