Vulnerabilities
All
My stack
Known exploited
Public exploit
Critical (9+)
Likely exploited (EPSS ≥ 50%)
Changed, 24 h
New, 24 h
Sort:
risk
modified
published
CVSS
EPSS
| CVE | CVSS | EPSS | Vendor, product | Description | Modified |
|---|---|---|---|---|---|
| CVE-2026-59801 | 9.3 critical | 3.0% | no CPE yet | 9Router through version 0.4.41 contains an unauthenticated access vulnerability that allows remote attackers to interact with provider management API endpoints by sending requests without any credentials due to missing authentication middleware in the Next.js API routes under src/app/api/providers/*. Attackers can enumerate, create, modify, or delete provider connections to expose partial credentials, OAuth tokens, and API keys, redirect AI traffic to attacker-controlled servers, or cause complete denial of service by deleting all provider connections. | Sat 24 Oct |
| CVE-2026-42076 | 9.8 critical | 2.0% | no CPE yet | Evolver is a GEP-powered self-evolving engine for AI agents. Prior to version 1.69.3, a command injection vulnerability in the _extractLLM() function allows attackers to execute arbitrary shell commands on the server. The function constructs a curl command using string concatenation and passes it to execSync() without proper sanitization, enabling remote code execution when the corpus parameter contains shell metacharacters. This issue has been patched in version 1.69.3. | Mon 26 Oct |
| CVE-2025-67506 | 9.8 critical | 2.0% | pipeshub | PipesHub is a fully extensible workplace AI platform for enterprise search and workflow automation. Versions prior to 0.1.0-beta expose POST /api/v1/record/buffer/convert through missing authentication. The endpoint accepts a file upload and converts it to PDF via LibreOffice by uploading payload to os.path.join(tmpdir, file.filename) without normalizing the filename. An attacker can submit a crafted filename containing ../ sequences to write arbitrary files anywhere the service account has permission, enabling remote file overwrite or planting malicious code. This issue is fixed in version 0.1.0-beta. | 14 Oct |
| CVE-2025-6984 | 7.5 high | 2.0% | no CPE yet | The langchain-ai/langchain project, specifically the EverNoteLoader component, is vulnerable to XML External Entity (XXE) attacks due to insecure XML parsing. The affected version is 0.3.63. The vulnerability arises from the use of etree.iterparse() without disabling external entity references, which can lead to sensitive information disclosure. An attacker could exploit this by crafting a malicious XML payload that references local files, potentially exposing sensitive data such as /etc/passwd. | 19 Oct |
| CVE-2026-73851 | 6.1 medium | 2.0% | no CPE yet | Kiota is an OpenAPI based HTTP Client code generator. Prior to 1.29.1 and 1.34.0, an attacker who controls or tampers with the OpenAPI description consumed by Kiota can supply a file reference that resolves outside the manifest package (e.g. ../../../../etc/passwd, an absolute path, or a file:// / http(s):// URI). When the generated manifest is deployed and consumed by an AI host, this can lead to inclusion or disclosure of files outside the intended package boundary. This vulnerability is fixed in 1.29.1 and 1.34.0. | Sat 24 Oct |
| CVE-2026-85889 | 10.0 critical | 1.0% | microsoft, azure ai foundry | Missing authentication for critical function in Azure AI Foundry allows an unauthorized attacker to elevate privileges over a network. | Mon 26 Oct |
| CVE-2026-35435 | 10.0 critical | 1.0% | microsoft, azure_ai_foundry | Improper access control in Azure AI Foundry M365 published agents allows an unauthorized attacker to elevate privileges over a network. | Mon 26 Oct |
| CVE-2026-67429 | 10.0 critical | 1.0% | no CPE yet | Flyto2 Core is an execution kernel for automation and AI-agent workflows. Prior to 2.26.6, image.download and related file-writing modules use caller-controlled output_dir instead of validate_path_with_env_config and its FLYTO_SANDBOX_DIR confinement, allowing attacker-controlled response bytes to be written to arbitrary filesystem paths the process can access. This issue is fixed in version 2.26.6. | Sat 24 Oct |
| CVE-2026-70352 | 10.0 critical | 1.0% | microsoft, azure_ai_language_authoring | Missing authentication for critical function in Azure AI Language allows an unauthorized attacker to elevate privileges over a network. | 14 Oct |
| CVE-2026-41512 | 9.9 critical | 1.0% | mozilla, 0din_scanner | ai-scanner is an AI model safety scanner built on NVIDIA garak. From version 1.0.0 to before version 1.4.1, there is a remote code execution vulnerability via JavaScript injection in `BrowserAutomation::PlaywrightService`. This issue has been patched in version 1.4.1. | Mon 26 Oct |
| CVE-2026-105697 | 9.9 critical | 1.0% | no CPE yet | Langflow is a tool for building and deploying AI-powered agents and workflows. Before Langflow 1.10.3, the MCP stdio transport launched whatever command / args a user put in an MCP server configuration, with no allowlist and (before 1.10.3) wrapped in bash -c "exec {command} ...". Any user able to reach the MCP server settings ("Settings → MCP Servers → Add MCP Server", POST/PATCH /api/v2/mcp/servers/{server_name}) or to build a flow with the MCP Tools component could add a "server" whose command is an arbitrary OS command (touch, rm -rf, a reverse shell, ...). The command runs on the Langflow host as the Langflow process user as soon as Langflow tries to connect to the server (listing servers, loading tools, running the flow) — even when the UI then reports that the stdio server failed to start. With the default LANGFLOW_AUTO_LOGIN=true, GET /api/v1/auto_login hands out a token without credentials, so on an exposed instance running the default configuration this is reachable without an account. AUTO_LOGIN is documented as a development-only setting; with it disabled, any authenticated (non-admin) user can exploit it. This issue is fixed in Langflow 1.10.3, langflow-base 0.10.3, and lfx 1.10.3. | 9 Oct |
| CVE-2026-40351 | 9.8 critical | 1.0% | fastgpt | FastGPT is an AI Agent building platform. In versions prior to 4.14.9.5, the password-based login endpoint uses TypeScript type assertion without runtime validation, allowing an unauthenticated attacker to pass a MongoDB query operator object (e.g., {"$ne": ""}) as the password field. This NoSQL injection bypasses the password check, enabling login as any user including the root administrator. This issue has been fixed in version 4.14.9.5. | Mon 26 Oct |
| CVE-2026-15982 | 9.8 critical | 1.0% | no CPE yet | The Aimogen Pro - All-in-One AI Content Writer, Editor, ChatBot & Automation Toolkit plugin for WordPress is vulnerable to Privilege Escalation in all versions up to, and including, 2.8.4. This is due to due to a missing capability check on the 'aiomatic_call_google_ai_function' function. This makes it possible for unauthenticated attackers to leverage the 'aimogen_wp_god_mode' tool to clear function blacklists and execute arbitrary PHP functions, such as creating administrator accounts. | Sat 24 Oct |
| CVE-2026-50755 | 9.8 critical | 1.0% | no CPE yet | An issue in DayuanJiang next-ai-draw-io 0.4.13 allows a remote attacker to obtain sensitive information via the X-Forwarded-For header value | Sat 24 Oct |
| CVE-2026-65581 | 9.8 critical | 1.0% | no CPE yet | Unauthenticated PHP Object Injection in AI ANN <= 1.29.0 versions. | Sat 24 Oct |
| CVE-2026-42380 | 9.8 critical | 1.0% | no CPE yet | Unauthenticated PHP Object Injection in AI Lab < 5.4.2 versions. | Sat 24 Oct |
| CVE-2026-68502 | 9.8 critical | 1.0% | no CPE yet | LazyOwn RedTeam/APT Framework is an AI-powered C2 and red-team operations framework. Prior to 0.2.154, LazyOwn's lazyc2.py registers an unauthenticated Socket.IO input event handler that dispatches data.get('value') to LazyOwnShell.one_cmd, reaching LazyOwnShell.do_cmd and subprocess.call(command, shell=True), allowing unauthenticated remote code execution in the C2 process. This issue is fixed in 0.2.154. | Fri 23 Oct |
| CVE-2026-68503 | 9.8 critical | 1.0% | no CPE yet | LazyOwn RedTeam/APT Framework is an AI-powered C2 and red-team operations framework. Prior to 0.2.154, LazyOwn ships default C2 credentials LazyOwn and LazyOwn in payload.json and core/payload_schema.py and passes them unchanged to lazyc2.py HTTP Basic authentication, allowing any network-reachable attacker who knows the defaults to authenticate to the C2 dashboard with operator-level access. This issue is fixed in 0.2.154. | Fri 23 Oct |
| CVE-2026-89492 | 9.8 critical | 1.0% | no CPE yet | In the Linux kernel, the following vulnerability has been resolved: ocfs2: validate directory-index entry counts when reading metadata ocfs2_validate_dx_leaf() and ocfs2_validate_dx_root() check the ECC and signature of an indexed-directory block before it reaches higher-level callers, but neither validator bounds the ocfs2_dx_entry_list counts against the capacity of the block that holds them. ocfs2_dx_dir_search() then walks for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) dx_entry = &entry_list->de_entries[i]; over de_num_used entries with no bounds check. entry_list is either dx_leaf->dl_list (from ocfs2_read_dx_leaf) or, for an inline root, dx_root->dr_entries. A crafted on-disk image can set de_num_used (and de_count, which is the __counted_by_le() bound of de_entries) to 0xffff and make the walk read far past the end of the 4KB metadata block, giving a slab out-of-bounds read reachable from any path lookup, stat() or open() on an indexed directory once the image is mounted. Commit 775c17386a6f ("ocfs2: validate dx_root extent list fields during block read") already bounds dr_list for the non-inline dx_root, but left the inline dr_entries path and the dx_leaf dl_list unchecked. Add the same read-time validation for both entry lists: de_count must equal the capacity of the block (ocfs2_dx_entries_per_leaf()/per_root()) and de_num_used must not exceed de_count, rejecting corrupted metadata with -EFSCORRUPTED before ocfs2_dx_dir_search() can walk an out-of-range entry array. de_count is always written as exactly the block capacity when a leaf or inline root is formatted, so the equality check does not reject any valid image. Found by 0sec automated security-research tooling (https://0sec.ai). | 21 Oct |
| CVE-2025-59944 | 9.8 critical | 1.0% | anysphere, cursor | Cursor is a code editor built for programming with AI. Versions 1.6.23 and below contain case-sensitive checks in the way Cursor IDE protects its sensitive files (e.g., */.cursor/mcp.json), which allows attackers to modify the content of these files through prompt injection and achieve remote code execution. A prompt injection can lead to full RCE through modifying sensitive files on case-insensitive fileystems. This issue is fixed in version 1.7. | 19 Oct |
| CVE-2026-74427 | 9.8 critical | 1.0% | no CPE yet | In the Linux kernel, the following vulnerability has been resolved: afs: Fix netns teardown to cancel the preallocation charger Fix the teardown of an afs network namespace to make sure it cancels the work item that keeps the preallocated rxrpc call/conn/peer queue charged before incoming calls are disabled (i.e. listen 0). Also, if net->live is false because the afs netns is being deleted, make afs_charge_preallocation() skip charging and make afs_rx_new_call() avoid requeuing the charger. (This was found by AI review). | 16 Oct |
| CVE-2026-105038 | 9.8 critical | 1.0% | no CPE yet | The EnergizeLab Eilik and EilikO FastAPI AI inference and profiling services expose AI model administration functions without authentication. An unauthenticated remote attacker can change AI model configuration and switch the model in use at runtime. | 16 Oct |
| CVE-2026-90920 | 9.8 critical | 1.0% | no CPE yet | IBM Maximo AI Service 9.1.0, and 9.2.0 could allow a remote attacker to execute arbitrary code due to improper control of code generation. | 13 Oct |
| CVE-2026-55447 | 9.6 critical | 1.0% | langflow | Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to 1.9.2, by controlling a files that are digested into the RAG, an attacker can direct the node to read any file on the file-system by absolute path. All components based on BaseFileComponent are vulnerable to the vulnerability. This includes Docling (DoclingInlineComponent), Docling Serve, DoclingRemoteComponent), Read File (FileComponent), NVIDIA Retriever Extraction (NvidiaIngestComponent), Video File (VideoFileComponent), and Unstructured API (UnstructuredComponent). This vulnerability is fixed in 1.9.2. | Sun 25 Oct |
| CVE-2026-12045 | 9.4 critical | 1.0% | pgadmin, pgadmin_4 | Read-only transaction bypass in the pgAdmin 4 AI Assistant allows an attacker who can influence database content that the assistant reads to execute arbitrary SQL with the privileges of the pgAdmin user's database role. The AI Assistant's execute_sql_query tool runs LLM-generated SQL inside a BEGIN TRANSACTION READ ONLY wrapper to prevent data modification. The LLM-supplied query was forwarded to the database driver without restriction to a single statement or to read-only verbs, so a multi-statement payload beginning with COMMIT, END, ROLLBACK, or ABORT terminated the read-only transaction and ran subsequent statements in autocommit mode. The trailing ROLLBACK then had no effect. Delivery is via prompt injection: an attacker who can write content into any object the AI Assistant may inspect (a row, a column value, a comment) can cause the LLM to emit the multi-statement payload as a tool call. With ordinary write privileges on the pgAdmin user's role the attacker can perform unauthorised data modification. When the pgAdmin user's role is a PostgreSQL superuser or holds pg_execute_server_program, the chain extends to remote code execution on the database server host via COPY ... TO PROGRAM. Fix validates the LLM-supplied query up front: it must parse to exactly one non-empty / non-comment statement whose leading real token (after stripping whitespace, comments, and punctuation) is one of SELECT, WITH, EXPLAIN, SHOW, VALUES, or TABLE. Transaction-control verbs, DML, DDL, CALL, COPY, DO, SET/RESET, and everything else are rejected before any database work happens. PostgreSQL's READ ONLY mode continues to backstop data-modifying CTEs, EXPLAIN ANALYZE on writes, and volatile side effects. This issue affects pgAdmin 4: from 9.13 before 9.16. | Sun 25 Oct |
| CVE-2026-26210 | 9.3 critical | 1.0% | kvcache-ai, ktransformers | KTransformers through 0.5.3 contains an unsafe deserialization vulnerability in the balance_serve backend mode where the scheduler RPC server binds a ZMQ ROUTER socket to all interfaces with no authentication and deserializes incoming messages using pickle.loads() without validation. Attackers can send a crafted pickle payload to the exposed ZMQ socket to execute arbitrary code on the server with the privileges of the ktransformers process. | Mon 26 Oct |
| CVE-2026-45261 | 9.3 critical | 1.0% | no CPE yet | GitButler is a modern Git-based version control interface for AI-powered workflows. Prior to 0.19.7, a emote code execution vulnerability exists in the Tauri-based GitButler desktop application. An attacker can inject a malicious link in a pull request body, which if clicked by the user allows for arbitrary script execution in the Tauri webview. Users that have not enabled forge integration are not at risk. This vulnerability is fixed in 0.19.7. | Mon 26 Oct |
| CVE-2026-59864 | 9.3 critical | 1.0% | no CPE yet | Kiota is an OpenAPI based HTTP Client code generator. Prior to 1.29.1 and 1.32.5, `kiota plugin add` and `kiota plugin generate` (with `-t APIPlugin`) emitted attacker-controlled static_template.file values from x-ai-adaptive-card and x-ai-capabilities into generated Microsoft 365 Copilot and Teams plugin manifests without path validation, allowing ../, absolute, rooted, UNC, Windows drive, or URI paths in response_semantics.static_template.file to cause path traversal or out-of-package file inclusion when the generated plugin was deployed. This issue is fixed in version 1.29.1 and 1.32.5. | Sat 24 Oct |
| CVE-2026-15091 | 9.3 critical | 1.0% | ibm, engineering_ai_hub | IBM Engineering AI Hub 1.0.0, 1.1.0, and 1.2.0 could allow a remote attacker to execute arbitrary scripts due to improper neutralization of input during web page generation. | Sat 24 Oct |
| CVE-2026-64622 | 9.3 critical | 1.0% | no CPE yet | Network-AI (npm: network-ai) versions 5.12.2 through 5.13.3 fail to apply the configured authorization check (checkAuth/secret) to the ApprovalInbox GET read routes, so even when an operator configures a secret, unauthenticated actors can access sensitive approval request details. The GET /approvals/?status=all, GET /approvals/:id, GET /approvals/stats, and GET /approvals/sse routes disclose full ApprovalEntry content including action/target shell-command strings, file paths, justifications, and risk levels. All responses also carry a hardcoded Access-Control-Allow-Origin: * header, enabling cross-origin disclosure from any website the operator visits. This is an incomplete fix for GHSA-mxjx-28vx-xjjj. | Sat 24 Oct |
| CVE-2026-75110 | 9.3 critical | 1.0% | no CPE yet | MemOS is a memory operating system for LLMs and AI agents. In deployments where authentication is enabled (AUTH_ENABLED=true) but the undocumented, defaultless INTERNAL_SERVICE_SECRET environment variable is unset, the is_internal_request() check in src/memos/api/middleware/auth.py fails open: os.getenv("INTERNAL_SERVICE_SECRET") returns None and a request omitting the X-Internal-Service header also yields None, so the comparison None == None evaluates true. The request is then treated as a trusted internal principal and granted scopes: ["all"]. As a result, an unauthenticated remote attacker can reach the admin API-key management endpoints to mint API keys for any user, enumerate keys, revoke keys, and generate a master key for persistent privileged access, as well as all data endpoints. | Fri 23 Oct |
| CVE-2026-111692 | 9.3 critical | 1.0% | no CPE yet | DataBuff ai-apm-web through 0.1.10 contains an authentication bypass vulnerability that allows unauthenticated attackers to forge admin session tokens by using the hardcoded HMAC secret in JwtTokenService. Attackers can sign a payload of admin and a future expiry with the default dev-secret to gain full administrative access to protected /webapi routes, including the Doris SQL query API. | 16 Oct |
| CVE-2026-111322 | 9.3 critical | 1.0% | no CPE yet | Archon 0.2.13 through 0.11.1 contains a missing authentication vulnerability that allows unauthenticated attackers to access the /api/* control API, which binds to 0.0.0.0 by default. Attackers with network access can send messages to the AI coding agent, which runs with bypassPermissions, to execute commands on the host as the Archon user. | 16 Oct |
| CVE-2026-20266 | 9.1 critical | 1.0% | splunk, ai_toolkit | In Splunk AI Toolkit versions below 5.7.4, a user who holds the "admin" Splunk role could execute arbitrary OS commands on the host running the Splunk Enterprise instance. The vulnerability is possible because of an unsafe shell execution pattern in the btool configuration helper, which constructs OS command strings from dynamic parameters without disabling shell interpretation. | Mon 26 Oct |
| CVE-2026-71479 | 9.1 critical | 1.0% | no CPE yet | New API is a large language mode (LLM) gateway and artificial intelligence (AI) asset management system. Prior to 1.0.0-rc.18, user-controlled image n, video seconds and duration, max_tokens, max_completion_tokens, maxOutputTokens, audio duration, and billing-expression quantities can overflow conversions in common/quota_math.go and related settlement paths, allowing a low-privileged account with positive balance or an active subscription to turn a negative charge into account credit and potentially drain upstream funds. This issue is fixed in version 1.0.0-rc.18. | Sat 24 Oct |
| CVE-2026-48039 | 9.1 critical | 1.0% | no CPE yet | Meta Ads MCP is a Model Context Protocol (MCP) server that lets AI assistants run Meta Ads. Prior to version 1.0.109, `AuthInjectionMiddleware.dispatch()` at `http_auth_integration.py:272` unconditionally forwards unauthenticated Streamable HTTP requests to downstream MCP tool handlers without issuing a `401` response, allowing any network-reachable caller to invoke MCP tools without authentication. When no per-request credential is present, tool handlers fall back to the `META_ACCESS_TOKEN` environment variable, and when the downstream Meta Graph API call fails, `api.py:263–269` serialises the raw `httpx` request URL—including the operator's `access_token` as a query parameter—into the JSON-RPC response body, delivering the credential to the unauthenticated caller. Version 1.0.109 fixes the issue. | Fri 23 Oct |
| CVE-2026-105032 | 9.1 critical | 1.0% | no CPE yet | The EnergizeLab Eilik and EilikO AI skill upload endpoint, served by the FastAPI AI inference and profiling services, does not require authentication and does not constrain uploaded file paths. An unauthenticated remote attacker can use path traversal to write arbitrary files, and thereby tamper with AI skill files used by Eilik and EilikO robots. | 16 Oct |
| CVE-2026-62674 | 9.0 critical | 1.0% | no CPE yet | Omnigent is an open-source AI agent framework and meta-harness for orchestrating coding agents. Prior to 0.3.0, PUT /sessions/{session_id}/agent checks LEVEL_EDIT permission for a session but does not reject a bound shared or template agent whose agent.session_id is None. An authenticated user with edit access to a session can replace that shared agent bundle through omnigent/server/routes/sessions.py, add a stdio MCP server, and cause later sessions that use the shared agent to launch an attacker-controlled command through omnigent/tools/mcp.py. The command executes with the Omnigent runner process permissions and can expose files, credentials, workspace data, internal services, and runner availability. This issue is fixed in version 0.3.0. | 14 Oct |
| CVE-2026-33654 | 8.9 high | 1.0% | nanobot | nanobot is a personal AI assistant. Prior to version 0.1.6, an indirect prompt injection vulnerability exists in the email channel processing module (`nanobot/channels/email.py`), allowing a remote, unauthenticated attacker to execute arbitrary LLM instructions (and subsequently, system tools) without any interaction from the bot owner. By sending an email containing malicious prompts to the bot's monitored email address, the bot automatically polls, ingests, and processes the email content as highly trusted input, fully bypassing channel isolation and resulting in a stealthy, zero-click attack. Version 0.1.6 patches the issue. | Mon 26 Oct |
| CVE-2026-1928 | 8.8 high | 1.0% | no CPE yet | The Rank Math SEO – AI SEO Tools to Dominate SEO Rankings plugin for WordPress is vulnerable to PHP Object Injection in all versions up to, and including, 1.0.276 via deserialization of untrusted input in the updateMeta REST API endpoint. This makes it possible forauthenticated attackers, with Contributor-level access and above, to inject a PHP Object. The additional presence of a POP chain allows attackers to achieve remote code execution by writing a webshell to the uploads directory. | Wed 28 Oct |
| CVE-2026-39981 | 8.8 high | 1.0% | agixt | AGiXT is a dynamic AI Agent Automation Platform. Prior to 1.9.2, the safe_join() function in the essential_abilities extension fails to validate that resolved file paths remain within the designated agent workspace. An authenticated attacker can use directory traversal sequences to read, write, or delete arbitrary files on the server hosting the AGiXT instance. This vulnerability is fixed in 1.9.2. | Mon 26 Oct |
| CVE-2026-41208 | 8.8 high | 1.0% | paperclip, paperclipai | Paperclip is a Node.js server and React UI that orchestrates a team of AI agents to run a business. Versions of @paperclipai/server prior to 2026.416.0 contain a privilege escalation vulnerability that allows an attacker with an Agent API key to execute arbitrary OS commands on the Paperclip server host. An attacker with an agent credential can escalate privileges from the agent runtime to the Paperclip server host. The vulnerability occurs because agents are allowed to update their own adapterConfig via the /agents/:id API endpoint. The configuration field adapterConfig.workspaceStrategy.provisionCommand is later executed by the server runtime. As a result, an attacker controlling an agent credential can inject arbitrary shell commands which are executed by the Paperclip server during workspace provisioning. This breaks the intended trust boundary between agent runtime configuration and server host execution, allowing a compromised or malicious agent to escalate privileges and run commands on the host system. This vulnerability allows remote code execution on the server host. @paperclipai/server version 2026.416.0 fixes the issue. | Mon 26 Oct |
| CVE-2026-13717 | 8.8 high | 1.0% | no CPE yet | A flaw was found in the Red Hat OpenShift AI (RHOAI) MaaS Gateway. Improper configuration of the Gateway in a model-serving context allows a standard user with low privileges to intercept, read, log, and alter all MaaS model traffic. This includes sensitive information such as access keys, input prompts, and outputs, leading to significant information disclosure and data tampering. | Mon 26 Oct |
| CVE-2026-32174 | 8.8 high | 1.0% | microsoft, azure_ai_bot_service | Improper authentication in Azure Bot Service allows an authorized attacker to elevate privileges over a network. | Mon 26 Oct |
| CVE-2026-54309 | 8.8 high | 1.0% | n8n | n8n is an open source workflow automation platform. Prior to 2.25.7 and 2.26.2, when @n8n/mcp-browser is run in HTTP transport mode, the MCP endpoint accepts session initialization and tool invocation requests without any authentication. Any network-reachable client, or any website visited by the user, can establish an MCP session and invoke browser-control tools. Where the n8n AI Browser Bridge extension is installed and a browser connection is active, an unauthenticated caller can access browser-control capabilities including navigation, JavaScript evaluation, and cookie and storage access against the user's real browser profile. This issue only affects instances where @n8n/mcp-browser is run with the HTTP transport (--transport http). This vulnerability is fixed in 2.25.7 and 2.26.2. | Mon 26 Oct |
| CVE-2026-54232 | 8.8 high | 1.0% | vllm | vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.1, the vLLM Dockerfile is vulnerable to a dependency confusion attack through the flashinfer-jit-cache package. The package is installed from a custom index (flashinfer.ai/whl/) using --extra-index-url, but the package name was not registered on PyPI, and UV_INDEX_STRATEGY="unsafe-best-match" is set globally. An attacker who registers flashinfer-jit-cache on PyPI with version 0.6.11.post2 can execute arbitrary code as root during the Docker build and backdoor every resulting container image, enabling exfiltration of all user prompts, API credentials, and model data from production vLLM deployments This vulnerability is fixed in 0.22.1. | Sun 25 Oct |
| CVE-2026-55249 | 8.8 high | 1.0% | rtkai, rtk-rewrite | @rtk-ai/rtk-rewrite transparently rewrites shell commands executed via OpenClaw's exec tool to their RTK equivalents. In 1.0.0, the @rtk-ai/rtk-rewrite OpenClaw plugin passes attacker-controlled input directly into a shell-backed execSync() template string without shell-safe escaping. JSON.stringify() wraps the value in double quotes and escapes inner double-quotes and backslashes, but leaves $() and backtick shell metacharacters untouched. Because execSync delegates execution to /bin/sh -c, the shell expands $(...) substitutions even inside double-quoted strings, causing the injected subcommand to execute before rtk is invoked. An attacker who can influence the exec tool's command parameter (e.g., via an LLM agent prompt or gateway/tool-call input) achieves arbitrary OS command execution with the privileges of the plugin/gateway process. | Sun 25 Oct |
| CVE-2026-62312 | 8.8 high | 1.0% | no CPE yet | 9Router is an AI router & token saver. Prior to 0.5.2, 9Router allows a remote authenticated attacker to achieve arbitrary code execution on the host operating system by combining a Host header bypass of localhost-only routes with unvalidated MCP plugin args passed to child_process.spawn(), allowing malicious custom plugins to execute commands through /api/mcp//sse. This issue is fixed in version 0.5.2. | Sat 24 Oct |
| CVE-2026-58195 | 8.8 high | 1.0% | no CPE yet | Agentic-Flow is an AI agent orchestration platform. Prior to 2.0.14, agentic-flow MCP server tools in src/mcp/standalone-stdio.ts, src/mcp/fastmcp/servers/claude-flow-sdk.ts, src/mcp/fastmcp/servers/stdio-full.ts, src/mcp/fastmcp/servers/http-streaming-updated.ts, src/mcp/fastmcp/servers/http-sse.ts, src/mcp/fastmcp/servers/poc-stdio.ts, src/mcp/fastmcp/tools/agent/{execute,list,parallel}.ts, src/mcp/fastmcp/tools/swarm/orchestrate.ts, and src/mcp/fastmcp/tools/hooks/pretrain.ts interpolated attacker-influenceable tool parameters such as agent, task, name, language, and agentdb directly into shell command strings passed to execSync(), allowing arbitrary OS command execution with the privileges of the MCP server user. This issue is fixed in version 2.0.14. | Sat 24 Oct |
| CVE-2026-56167 | 8.8 high | 1.0% | microsoft, azure_ai_search | Server-side request forgery (ssrf) in Azure AI Search allows an authorized attacker to elevate privileges over a network. | Sat 24 Oct |
Download these 704 as CSV · subscribe to this list.
EPSS scores: 2026-10-28. NVD cursor: 2026-10-29 00:22. The News page shows only notable CVEs (KEV, your stack, new criticals, EPSS over 50%, or active exploitation per SSVC); this table has all of them.