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-2026-62674 | 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-2026-39981 | 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-2026-55249 | 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-2026-57301 | 8.8 high | 1.0% | jenkins, official_owasp_zap | Jenkins OWASP ZAP Plugin 1.0.7 and earlier performs build operations on the Jenkins controller rather than the assigned agent, allowing attackers with Item/Configure permission to execute arbitrary code on the Jenkins controller. | Sun 25 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-2025-61591 | 8.8 high | 1.0% | anysphere, cursor | Cursor is a code editor built for programming with AI. In versions 1.7 and below, when MCP uses OAuth authentication with an untrusted MCP server, an attacker can impersonate a malicious MCP server and return crafted, maliciously injected commands during the interaction process, leading to command injection and potential remote code execution. If chained with an untrusted MCP service via OAuth, this command injection vulnerability could allow arbitrary code execution on the host by the agent. This can then be used to directly compromise the system by executing malicious commands with full user privileges. This issue does not currently have a fixed release version, but there is a patch, 2025.09.17-25b418f. | 19 Oct |
| CVE-2026-50112 | 8.8 high | 1.0% | apache, cloudstack | SSRF via Metalink Mirror URL Resolution: An authenticated tenant can register a template pointing to an attacker-controlled metalink file containing internal targets. The Secondary Storage VM will retrieve the data and persist it as a template file, which can later be downloaded through normal APIs. RCE on KVM hypervisor via NFS, Metalink files with/without Direct Downloads: An authenticated CloudStack tenant holding the default User role can execute arbitrary shell commands as root on the KVM hypervisor host that runs other tenants' VMs. This is cross-tenant root on the underlying compute, reachable via the public CloudStack API. When a User registers a VM template with directDownload=true and a URL pointing to a .metalink file, the management server fetches the metalink XML and dispatches download to the KVM agent. Inner URLs inside the metalink XML are never re-validated against the scheme allowlist. These issues affect Apache CloudStack: from 4.14.0.0 through 4.20.3.0 and from 4.21.0.0 through 4.22.1.0. Users are recommended to upgrade to version 4.20.3.1 or 4.22.1.1 or later, which fixes the issue. | 16 Oct |
| CVE-2026-63046 | 8.8 high | 1.0% | apache, inlong | Improper Neutralization of Argument Delimiters in a Command ('Argument Injection') vulnerability in Apache InLong. Agent Installer's ModuleManager executes arbitrary shell commands via ExcuteLinux.exeCmd() with no filtering or whitelist validation. This issue affects Apache InLong: from 2.0.0 before 2.4.0. Users are advised to upgrade to Apache InLong's 2.4.0 or cherry-pick [1]/[2] to solve it. [1] https://github.com/apache/inlong/pull/12151 . [2] https://github.com/apache/inlong/pull/12155 . | 16 Oct |
| CVE-2025-62550 | 8.8 high | 1.0% | microsoft, azure_monitor_agent | Out-of-bounds write in Azure Monitor Agent allows an authorized attacker to execute code over a network. | 14 Oct |
| CVE-2026-62675 | 8.8 high | 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, multipart POST /v1/sessions accepts an authenticated user's agent bundle and omnigent/server/bundles.py validate_agent_bundle does not reject a tools..callable dotted Python path. omnigent/runner/tool_dispatch.py _resolve_spec_callable imports the specified module and _execute_spec_callable_tool invokes the resolved function, allowing a bundle to select subprocess.check_output and execute a local command with the runner process permissions. This can expose runner files, environment variables, credentials, workspace data, internal services, and availability without administrator access. This issue is fixed in version 0.3.0. | 14 Oct |
| CVE-2026-62677 | 8.8 high | 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, an authenticated user can upload a session-scoped agent bundle with an absolute or traversal-containing os_env.cwd value because omnigent/spec/parser.py stores the value verbatim and omnigent/spec/validator.py does not constrain it. On a runner where OMNIGENT_RUNNER_WORKSPACE is unset, omnigent/runner/resource_registry.py preserves the attacker-controlled path and omnigent/inner/os_env.py uses the resolved path as the environment root and copytree source. The _assert_within_cwd check then treats that attacker-selected root as trusted, allowing sys_os_read, write, edit, and shell tools to access runner files and environment secrets outside the intended workspace. This issue is fixed in version 0.3.0. | 14 Oct |
| CVE-2026-42856 | 8.7 high | 1.0% | no CPE yet | Network-AI is a TypeScript/Node.js multi-agent orchestrator. Prior to 5.1.3, the MCP HTTP transport accepts JSON-RPC tools/call requests with no authentication, session, origin, or token check, and dispatches them directly to the orchestrator's tool registry. The default bind address is 0.0.0.0. As a result, any party with network reachability to the service can enumerate and invoke privileged management tools. This vulnerability is fixed in 5.1.3. | Mon 26 Oct |
| CVE-2026-53869 | 8.7 high | 1.0% | no CPE yet | Hermes Agent before 0.16.0 contains a DNS rebinding vulnerability in WebSocket endpoints that allows remote attackers to bypass Host and Origin validation. FastAPI HTTP middleware does not execute for WebSocket upgrade requests on /api/pty, /api/ws, /api/pub, and /api/events endpoints, enabling attackers to exploit DNS rebinding and inject malicious commands or read terminal output. | Mon 26 Oct |
| CVE-2026-73298 | 8.7 high | 1.0% | no CPE yet | The Microsoft Container Migration Solution Accelerator is a multi-service application that provides a multi-agent, AI-driven migration solution for moving container service configurations to Azure Kubernetes Service. In version 2.1.2 and earlier, a security vulnerability was identified in the Container Migration Solution Accelerator, specifically an authenticated IDOR (Insecure Direct Object Reference) that allows users to read, write, and delete processes belonging to other authenticated users. The issue affects multiple API endpoints, where ownership checks are missing, enabling unauthorized access and modification of migration data across users within the same organization. The vulnerability is present in both process and file management APIs, and the application relies on Entra ID authentication but lacks proper authorization controls between users. Authenticated users are able to access, modify, and delete processes and files belonging to other users without proper authorization checks. | Sat 24 Oct |
| CVE-2026-109440 | 8.7 high | 1.0% | no CPE yet | LangAlpha through 2026.09.07 contains an indirect prompt injection vulnerability that allows remote attackers to execute shell commands or Python by planting instructions in web content retrieved by the agent. Because Bash and ExecuteCode lack an allow-list and human-in-the-loop approval, attackers can run code in the workspace sandbox and exfiltrate vault secrets from _internal/.vault_secrets.json. | 22 Oct |
| CVE-2026-114107 | 8.7 high | 1.0% | no CPE yet | HKUDS Vibe-Trading from 0.1.0 before 0.1.7 contains an indirect prompt injection vulnerability that allows remote attackers to execute shell commands as root by planting instructions in documents or web pages the agent processes. BashTool and BackgroundRunTool pass model-emitted commands to subprocess.run with shell=True and no allow-list or approval step, and read_url fetches arbitrary URLs, enabling command execution and server-side request forgery. | 22 Oct |
| CVE-2026-109443 | 8.7 high | 1.0% | no CPE yet | JoySafeter through 1.0.0 contains a remote code execution vulnerability that allows authenticated users to run operating system commands via an unsandboxed execute tool in Copilot standard mode. Attackers can self-register, select standard mode in the WebSocket chat.start extension, and have the agent invoke subprocess.run with shell=True on the host to access other tenants' data and server secrets. | 20 Oct |
| CVE-2026-77926 | 8.7 high | 1.0% | no CPE yet | Aix-DB 1.2.4 contains a remote code execution vulnerability that allows authenticated non-admin users to run operating system commands through the ChatBI common agent's ungated deepagents execute tool. Attackers can send crafted chat input to POST /dify/get_answer or upload documents with malicious text to run host commands, exposing datasource credentials and model API keys. | 20 Oct |
| CVE-2026-86831 | 8.7 high | 1.0% | no CPE yet | Improper validation of pod identifier uniqueness in aws-network-policy-agent in Amazon EKS Network Policy Agent before v1.4.0 might allow an authenticated remote user to bypass NetworkPolicy enforcement on co-located pods in other namespaces via crafted pod and namespace names that produce pod identifier collisions. To remediate this issue, users should upgrade to Amazon EKS Network Policy Agent 1.4.0 or later and Amazon VPC CNI Managed Add-on v1.22.4 or later (which includes Network Policy Agent v1.4.0). | 20 Oct |
| CVE-2026-73040 | 8.7 high | 1.0% | no CPE yet | Dockge validates a stack name only on the write path. In backend/stack.ts the allow-list check in validate(), which requires the name to match ^[a-z0-9_-]+$, is reached from save() alone, while the path getter returns path.join(this.server.stacksDir, this.name) and Stack.getStack builds path.join(server.stacksDir, stackName) with no check. The socket handlers in backend/agent-socket-handlers/docker-socket-handler.ts confirm the caller is logged in and that the name is a string, then pass it straight to Stack.getStack, so a name containing traversal sequences resolves outside the managed stacks directory. An authenticated user can therefore read the composeENV and composeYAML values of any directory the server process can reach, which discloses the secrets in that directory's .env or Compose file, and can invoke delete(), which runs docker compose down and then fsAsync.rm on the traversed path with recursive and force set, removing that directory. Disclosure is limited to files named .env or an accepted Compose filename, and deletion requires the target directory to hold a valid Compose file so that docker compose down exits successfully. Dockge commonly runs as root with access to the Docker socket, so the reachable set includes unrelated applications on the host. Instances configured with disableAuth, a supported option that logs the caller in as admin automatically, expose both operations without authentication. | 18 Oct |
| CVE-2026-109429 | 8.7 high | 1.0% | no CPE yet | Mistral Vibe before 2.25.4 contains a command permission gate bypass vulnerability that allows attackers to execute arbitrary commands by hiding a command substitution inside a single-quoted variable value that tree-sitter-bash static parsing cannot detect. Attackers can craft input such as embedding a command substitution within a single-quoted variable and re-expanding it with the ${var@P} prompt-transform operator so bash's promptvars feature executes the hidden command at runtime while only the trailing allowlisted command is classified, resulting in silent arbitrary command execution with the privileges of the agent's terminal session. | 16 Oct |
| CVE-2026-56075 | 8.7 high | 1.0% | no CPE yet | PraisonAI before 4.5.128 contains an arbitrary shell command execution vulnerability where the UI modules hardcode approval_mode to auto, overriding administrator configuration from PRAISON_APPROVAL_MODE environment variable. Authenticated attackers can instruct the LLM agent to execute arbitrary shell commands via subprocess.run with shell=True, bypassing the manual approval gate and insufficient command sanitization blocklists. | 8 Oct |
| CVE-2026-14933 | 8.6 high | 1.0% | no CPE yet | Cross-site scripting in the embedded co-browsing feature in Unblu inc. Unblu Spark 7.0 through 7.57.7 and 8.0 through 8.39.2, on on-premises and cloud deployments, allows an unauthenticated remote attacker in the visitor role to execute arbitrary JavaScript in the browser of an agent taking part in the attacker's co-browsing session, via content of the co-browsed page that the agent's browser renders as HTML. The JavaScript runs in the context and under the identity of the agent's session in the Agent Desk or the Single Conversation Desk. Exploitation requires an agent to join an embedded co-browsing session with the attacker. Mobile co-apping isn't affected. | 20 Oct |
| CVE-2025-71421 | 8.6 high | 1.0% | no CPE yet | UVdesk core-framework before 1.1.7 contains an improper privilege management vulnerability in the editAgent endpoint that allows agents with agent-management privilege to escalate their own role to administrator. Attackers can submit their own account identifier with a role parameter set to ROLE_ADMIN to gain full administrative control over agents, tickets, and mail configuration. | 14 Oct |
| CVE-2026-56076 | 8.6 high | 1.0% | no CPE yet | PraisonAI before 1.5.128 contains a cross-origin agent execution vulnerability in the AGUI endpoint that allows remote attackers to trigger arbitrary agent execution. The POST /agui endpoint lacks authentication and hardcodes Access-Control-Allow-Origin: * headers, combined with Starlette's Content-Type-agnostic JSON parsing, enabling attackers to bypass CORS preflight checks via simple requests and exfiltrate sensitive agent responses including tool execution results and environment data. | 8 Oct |
| CVE-2026-117567 | 8.5 high | 1.0% | no CPE yet | esp-mcp through commit 28afddf contains an OS command injection vulnerability that allows attackers to run arbitrary commands by supplying shell metacharacters in MCP tool parameters. Attackers influencing agent tool calls through prompt injection can inject metacharacters into target, port, project_path, or project_name values passed to asyncio.create_subprocess_shell, executing commands as the MCP server user. | Tue 27 Oct |
| CVE-2026-76832 | 8.5 high | 1.0% | agno agi, agno | Agno's PythonTools in libs/agno/agno/tools/python.py contains a path traversal vulnerability that allows attackers to read, write, or execute arbitrary files by supplying parent-directory traversal sequences in the file_name argument passed to read_file, save_to_file, or run_python_file tool actions. Attackers can inject traversal sequences such as '../../../../../../etc/passwd' through direct tool invocation or via prompt injection embedded in agent-processed content to escape the intended base_dir boundary and achieve arbitrary file read, arbitrary file write, or arbitrary Python code execution within the process user's authority. | Tue 27 Oct |
| CVE-2026-44688 | 8.4 high | 1.0% | eclipse, theia | In Eclipse Theia versions prior to 1.71.0, the AI chat agent processed workspace file and directory names as part of its prompt context without distinguishing them from system instructions. An attacker could craft a malicious repository with adversarial directory or file names that, when analyzed by the AI agent, would cause the agent to follow attacker-controlled instructions (indirect prompt injection). Combined with other AI chat features available in untrusted workspaces, this enabled attack chains leading to data exfiltration via Markdown image rendering or arbitrary command execution via task definitions. | Mon 26 Oct |
| CVE-2026-46580 | 8.4 high | 1.0% | eclipse, theia | In Eclipse Theia versions prior to 1.71.0, files matching the pattern .prompts/*.prompttemplate in a workspace were automatically loaded and could override or extend the AI agent's system prompts. An attacker could craft a malicious repository containing prompt template files that, when the workspace was opened in Theia, replaced the AI's system instructions with attacker-controlled content (indirect prompt injection). Combined with other AI chat features available in untrusted workspaces, this enabled attack chains leading to data exfiltration via Markdown image rendering or arbitrary command execution via task definitions. | Mon 26 Oct |
| CVE-2026-47398 | 8.1 high | 1.0% | no CPE yet | PraisonAI is a multi-agent teams system. The v4.6.32 chokepoint refactor (which patched CVE-2026-44334 / GHSA-xcmw-grxf-wjhj) added the PRAISONAI_ALLOW_LOCAL_TOOLS env-var gate to the tool_override.py sinks. However, two additional spec.loader.exec_module call sites in praisonai/agents_generator.py were missed and remain completely unguarded in versions prior to 4.6.40. Both functions accept a module_path parameter sourced from YAML configuration and execute it without validation, signature checking, or the env-var gate. Version 4.6.40 fixes the issue. | Sat 24 Oct |
| CVE-2026-47412 | 8.1 high | 1.0% | no CPE yet | PraisonAI Platform is the platform layer for the PraisonAI multi-agent teams system. Versions prior to 0.1.4 have an authorization bypass enabling destructive action. The `DELETE /workspaces/{workspace_id}` endpoint is gated only by `require_workspace_member(workspace_id)` (default `min_role="member"`). Any member of the workspace can issue a single DELETE to wipe the entire workspace, including every project, issue, comment, agent, label, and member record (cascading via the foreign-key relationships). There is no owner-role gate, no confirmation token, no soft-delete window, no recovery path. PraisonAI Platform version 0.1.4 patches the issue. | Sat 24 Oct |
| CVE-2025-65946 | 8.1 high | 1.0% | roocode, roo_code | Roo Code is an AI-powered autonomous coding agent that lives in users' editors. Prior to version 3.26.7, Due to an error in validation it was possible for Roo to automatically execute commands that did not match the allow list prefixes. This issue has been patched in version 3.26.7. | 18 Oct |
| CVE-2026-69435 | 8.1 high | 1.0% | microsoft, azure_sre_agent | Missing authorization in Azure SRE Agent allows an authorized attacker to elevate privileges over a network. | 15 Oct |
| CVE-2025-34207 | 7.9 high | 1.0% | vasion, virtual_appliance_application, virtual_appliance_host | Vasion Print (formerly PrinterLogic) Virtual Appliance Host prior to 22.0.1049 and Application prior to 20.0.2786 (VA and SaaS deployments) configure the SSH client within Docker instances with the following options: `UserKnownHostsFile=/dev/null`, `StrictHostKeyChecking=no`, and `ForwardAgent yes`. These settings disable verification of the remote host’s SSH key and automatically forward the developer’s SSH‑agent to any host that matches the configured wildcard patterns. As a result, an attacker who can reach a single compromised container can cause the container to connect to a malicious SSH server, capture the forwarded private keys, and use those keys for unrestricted lateral movement across the environment. This vulnerability has been identified by the vendor as: V-2024-027 — Insecure Secure Shell (SSH) Configuration. | 18 Oct |
| CVE-2025-58724 | 7.8 high | 1.0% | microsoft, azure_connected_machine_agent | Improper access control in Azure Connected Machine Agent allows an authorized attacker to elevate privileges locally. | 8 Oct |
| CVE-2025-59494 | 7.8 high | 1.0% | microsoft, azure_monitor_agent | Improper access control in Azure Monitor Agent allows an authorized attacker to elevate privileges locally. | 8 Oct |
| CVE-2026-68746 | 7.7 high | 1.0% | livebook | Not Failing Securely ('Failing Open') vulnerability in livebook-dev livebook allows an unauthenticated network client to obtain full access to a Livebook server that enforces identity through Livebook Teams. A Livebook Agent or App Server connected to Livebook Teams caches the identifier of the deployment group it belongs to, and resolves that identifier against a locally cached list of deployment groups on every request in order to decide whether Teams identity enforcement is active. Livebook.Hubs.TeamClient.handle_call/3 in lib/livebook/hubs/team_client.ex does not distinguish a deployment group that could not be resolved from one that was resolved with identity enforcement switched off: the clause matches only the case where a group was found with enforcement enabled, and falls through to a catch-all that reports enforcement as switched off for everything else. The two neighbouring functions that decide user and application access resolve the same identifier and treat the same unresolved result as a denial. When the identity status is reported as switched off, Livebook.ZTA.LivebookTeams.authenticate/3 in lib/livebook/zta/livebook_teams.ex returns empty identity metadata and allows the request to continue instead of halting it. LivebookWeb.UserPlug.build_current_user/3 merges that empty metadata into a newly built user, whose access type defaults to full access, and LivebookWeb.AuthPlug.authorized?/1 grants access to any user holding full access. The cached identifier becomes unresolvable when the deployment group it refers to is deleted while the agent is not connected to receive the change, most concretely when a deployment group is deleted during the window in which an agent is disconnected or reconnecting. The client removes the group from its cached list without clearing the identifier that refers to it. Any client able to reach the affected server over the network is then granted the same access as a fully privileged member of the organisation, including the ability to read notebooks and configured secrets, execute code on the server's runtime, and disrupt its operation. This issue affects livebook: from 0.19.7 before 0.19.9. | Sat 24 Oct |
| CVE-2026-109560 | 7.7 high | 1.0% | no CPE yet | Continue CLI (@continuedev/cli) through 1.5.47 contains a permission bypass vulnerability that allows attackers to run unapproved commands by chaining them after allowed Bash prefix rules. Attackers can inject instructions into repository files or fetched pages so the agent appends commands after ;, && or | to execute code as the user. | 16 Oct |
| CVE-2026-43003 | 7.5 high | 1.0% | openstack, ironic_python_agent | An issue was discovered in OpenStack ironic-python-agent 1.0.0 through 11.5.0. Ironic Python Agent (IPA) sometimes executes grub-install from within a chroot of the deployed partition image, leading to code execution in the case of a malicious image. | Mon 26 Oct |
| CVE-2026-7461 | 7.5 high | 1.0% | amazon, amazon_ecs_container_agent | Improper neutralization of inputs used in an OS command in the FSx Windows File Server volume mounting component in Amazon ECS Agent on Windows before version 1.103.0 might allow a remote authenticated threat actor to execute shell commands with SYSTEM privileges on the underlying host via a specially crafted username field in an ECS task definition. This issue requires permissions to register ECS task definitions or write to the Secrets Manager or SSM Parameter Store credentials used by the FSx volume configuration. To remediate this issue, users should upgrade to version 1.103.0. | Mon 26 Oct |
| CVE-2026-18733 | 7.5 high | 1.0% | no CPE yet | A prompt injection vulnerability in the shell tool in Amazon Strands Agents Tools before 0.8.0 might allow remote actors to execute arbitrary operating system commands on the agent's host via a crafted prompt that sets the non_interactive parameter to true, bypassing the human consent gate. To remediate this issue, users should upgrade to version 0.8.0. | Fri 23 Oct |
| CVE-2025-61590 | 7.5 high | 1.0% | anysphere, cursor | Cursor is a code editor built for programming with AI. Versions 1.6 and below are vulnerable to Remote Code Execution (RCE) attacks through Visual Studio Code Workspaces. Workspaces allow users to open more than a single folder and save specific settings (pretty similar to .vscode/settings.json) for the folders / project. An untitled workspace is automatically created by VS Code (untitled.code-workspace), which contains all the folders and workspace settings from the user's current session, opening up an entire new attack vector if the user has a .code-workspace file in path (either untitled created automatically or a saved one). If an attacker is able to hijack the chat context of the victim (such as via a compromised MCP server), they can use prompt injection to make the Cursor Agent write into this file and modify the workspace. This leads to a bypass of CVE-2025-54130 which can lead to RCE by writing to the settings section. This issue is fixed in version 1.7. | 19 Oct |
| CVE-2025-47913 | 7.5 high | 1.0% | go, ssh | SSH clients receiving SSH_AGENT_SUCCESS when expecting a typed response will panic and cause early termination of the client process. | 18 Oct |
| CVE-2026-63043 | 7.5 high | 1.0% | apache, inlong | Relative Path Traversal vulnerability in Apache InLong. Arbitrary file read from the Agent host filesystem. This issue affects Apache InLong: from 2.0.0 before 2.4.0. Users are advised to upgrade to Apache InLong's 2.4.0 or cherry-pick [1] to solve it. [1] https://github.com/apache/inlong/pull/12146 . | 18 Oct |
| CVE-2026-105791 | 7.5 high | 1.0% | no CPE yet | Microsoft UFO is an open-source framework for intelligent automation across devices and platforms. Prior to 3.0.9, the run_shell tool in the CommandLineExecutor component of ufo/client/mcp/local_servers/cli_mcp_server.py validates only the first token of the bash_command parameter and permits explorer.exe. On Windows, explorer.exe delegates its following path argument to ShellExecute, so an attacker-influenced agent call can launch an arbitrary executable or script as the desktop user even though the subprocess uses shell=False. Exploitation depends on a user running an affected agent workflow and on inducing the tool call, but successful execution can access or modify that user's files, tokens, and sessions. This issue is fixed in version 3.0.9. | 9 Oct |
| CVE-2026-45539 | 7.4 high | 1.0% | no CPE yet | Microsoft APM is an open-source, community-driven dependency manager for AI agents. From 0.5.4 to 0.12.4, two primitive integrators in apm-cli enumerate package files with bare Path.glob() / Path.rglob() calls and read each match with Path.read_text(), transparently following symbolic links. A symlink committed inside a remote APM dependency under .apm/prompts/<x>.prompt.md or .apm/agents/<x>.agent.md is preserved verbatim into apm_modules/ on clone and then dereferenced during integration, with the resolved content written as a regular file into the project's deploy directories. The package content_hash, the pre-deploy SecurityGate scan, and apm audit do not flag this. The deploy roots are not added to the auto-generated .gitignore, so the resulting files are staged by git add by default. This vulnerability is fixed in 0.13.0. | Mon 26 Oct |
| CVE-2023-0361 | 7.4 high | 1.0% | gnu, redhat, gnutls, enterprise_linux, debian_linux | A timing side-channel in the handling of RSA ClientKeyExchange messages was discovered in GnuTLS. This side-channel can be sufficient to recover the key encrypted in the RSA ciphertext across a network in a Bleichenbacher style attack. To achieve a successful decryption the attacker would need to send a large amount of specially crafted messages to the vulnerable server. By recovering the secret from the ClientKeyExchange message, the attacker would be able to decrypt the application data exchanged over that connection. | 14 Oct |
| CVE-2026-96893 | 7.3 high | 1.0% | gitlab | GitLab has remediated an issue in GitLab EE affecting all versions from 19.1 before 19.2.8, 19.3 before 19.3.5, and 19.4 before 19.4.2 that, under certain conditions, could have allowed an authenticated user to inject specially crafted content into AI agent instructions, potentially causing unintended command execution attributed to other users, due to improper neutralization of special elements in AI conversation context construction. | 14 Oct |
| CVE-2026-58170 | 7.2 high | 1.0% | no CPE yet | Vibe-Trading before 0.1.10 builds the proposal file path by joining a caller-supplied proposal identifier onto the broker proposals directory without sanitization (agent/src/live/mandate/commit.py). A proposal identifier containing path traversal sequences causes the application to load an attacker-controlled JSON file as an authoritative live trading mandate. Combined with the file upload endpoint, an admitted caller can write a JSON file to a known location and traverse to it, and because the ceilings validation is skipped when ceilings are absent, the attacker fully controls the committed mandate. | Sun 25 Oct |
Download these 793 as CSV · subscribe to this list.
EPSS scores: 2026-10-29. 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.