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-2026-47414 | 7.6 high | 0.00% | no CPE yet | PraisonAI Platform is the platform layer for the PraisonAI multi-agent teams system. Versions prior to 0.1.4 have an Insecure Direct Object Reference. Five label endpoints — `PATCH /workspaces/{workspace_id}/labels/{label_id}`, `DELETE .../labels/{label_id}`, `POST .../issues/{issue_id}/labels/{label_id}`, `DELETE .../issues/{issue_id}/labels/{label_id}`, `GET .../issues/{issue_id}/labels` — gate access on `require_workspace_member(workspace_id)` only and pass URL-supplied `label_id` and `issue_id` straight through to `LabelService` without verifying either belongs to the workspace. PraisonAI Platform version 0.1.4 patches the issue. | Sat 24 Oct |
| CVE-2025-33203 | 7.6 high | 0.00% | no CPE yet | NVIDIA NeMo Agent Toolkit UI for Web contains a vulnerability in the chat API endpoint where an attacker may cause a Server-Side Request Forgery. A successful exploit of this vulnerability may lead to information disclosure and denial of service. | 18 Oct |
| CVE-2026-54640 | 7.6 high | 0.00% | no CPE yet | OpenRemote is an open-source internet-of-things platform. Prior to 1.24.2, the KNXProtocol.startAssetImport method in agent/src/main/java/org/openremote/agent/protocol/knx/KNXProtocol.java processes 0.xml from an authenticated user's uploaded ETS project ZIP with Saxon TransformerFactoryImpl and XMLInputFactory parsers that permit external entity resolution. This incomplete remediation of CVE-2026-40882 leaves the KNX handler unprotected while the parallel Velbus handler is hardened, allowing a regular realm user to read files accessible to the JVM process or cause server-side requests. This issue is fixed in version 1.24.2. | 16 Oct |
| CVE-2025-71427 | 7.6 high | 0.00% | no CPE yet | Office-PowerPoint-MCP-Server through 2.0.7 contains a path traversal vulnerability that allows MCP callers to write and read files outside the working directory by supplying absolute paths or ../ sequences. Attackers can steer an AI agent via prompt injection to abuse save_presentation, open_presentation, or manage_image output_path to overwrite any server-writable file or load external files. | 14 Oct |
| CVE-2026-107295 | 7.6 high | 0.00% | pydantic, pydantic_ai | Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 1.34.0 until 1.107.4 and 2.28.0, the Agent.to_web() and clai web development chat endpoint has missing request content-type validation. A website visited by a developer can submit a browser-compatible request to a loopback-hosted chat server, causing the served agent to run and execute tools with the privileges and credentials of the local process; client-relayed approval decisions also leave requires_approval=True tools exposed. Binding to localhost does not prevent a browser page from reaching the loopback address. This issue is fixed in versions 1.107.4 and 2.28.0. | 13 Oct |
| CVE-2026-40116 | 7.5 high | 0.00% | praison, praisonai | PraisonAI is a multi-agent teams system. Prior to 4.5.128, the /media-stream WebSocket endpoint in PraisonAI's call module accepts connections from any client without authentication or Twilio signature validation. Each connection opens an authenticated session to OpenAI's Realtime API using the server's API key. There are no limits on concurrent connections, message rate, or message size, allowing an unauthenticated attacker to exhaust server resources and drain the victim's OpenAI API credits. This vulnerability is fixed in 4.5.128. | Mon 26 Oct |
| CVE-2026-40117 | 7.5 high | 0.00% | praison, praisonaiagents | PraisonAIAgents is a multi-agent teams system. Prior to 1.5.128, read_skill_file() in skill_tools.py allows reading arbitrary files from the filesystem by accepting an unrestricted skill_path parameter. Unlike file_tools.read_file which enforces workspace boundary confinement, and unlike run_skill_script which requires critical-level approval, read_skill_file has neither protection. An agent influenced by prompt injection can exfiltrate sensitive files without triggering any approval prompt. This vulnerability is fixed in 1.5.128. | Mon 26 Oct |
| CVE-2026-40115 | 7.5 high | 0.00% | praison, praisonai | PraisonAI is a multi-agent teams system. Prior to 4.5.128, the WSGI-based recipe registry server (server.py) reads the entire HTTP request body into memory based on the client-supplied Content-Length header with no upper bound. Combined with authentication being disabled by default (no token configured), any local process can send arbitrarily large POST requests to exhaust server memory and cause a denial of service. The Starlette-based server (serve.py) has RequestSizeLimitMiddleware with a 10MB limit, but the WSGI server lacks any equivalent protection. This vulnerability is fixed in 4.5.128. | Mon 26 Oct |
| CVE-2026-105626 | 7.5 high | 0.00% | no CPE yet | HashiCorp Consul and Consul Enterprise are vulnerable to a denial of service in the HTTP API that may allow an unauthenticated attacker to exhaust memory on a Consul agent. This vulnerability (CVE-2026-105626) is fixed in Consul Community Edition 2.1.0 and Consul Enterprise 2.1.0 and 1.21.19. | Mon 26 Oct |
| CVE-2026-105765 | 7.5 high | 0.00% | no CPE yet | HashiCorp Consul and Consul Enterprise 1.17.0 up to 2.0.4 are vulnerable to an authorization bypass in the agent's service mesh leaf certificate API that may allow an attacker without a valid ACL token to retrieve a cached service leaf certificate and its private key. This vulnerability (CVE-2026-105765) is fixed in Consul Community Edition 2.1.0 and Consul Enterprise 2.1.0 and 1.21.19. | Mon 26 Oct |
| CVE-2026-109331 | 7.5 high | 0.00% | no CPE yet | Network-AI from 5.14.0 before 5.15.2 contains a security control bypass vulnerability in the ClaudeHookBridge PreToolUse hook, which matches deny patterns only against the first present candidate field. A prompt-injected agent can place dangerous payloads in fields like Write content or an MCP script parameter beside a benign file_path or command, evading denyPatterns to write web shells or execute scripts. | Mon 26 Oct |
| CVE-2026-8360 | 7.5 high | 0.00% | no CPE yet | Function calls to WOSCommonUtil.dll!WOSSysInfoGetDeviceInterface() in various DLLs (i.e., WOSProfileMgrModule.dll, WOSWebDavModule.dll) can return a NULL pointer (i.e., when no user is logged into the Triofox Server Agent Management Console). The returned NULL pointer is not checked before being dereferenced. | Mon 26 Oct |
| CVE-2026-65644 | 7.5 high | 0.00% | rocket.chat | Rocket.Chat in versions before 8.8.0, 8.7.1, 8.6.2, 8.5.3, 8.4.6, 8.3.8, 8.2.8, 8.1.8, and 7.10.15 has a REST API endpoint POST /api/v1/livechat/visitor that accepts an unauthenticated, unsanitized name field for Livechat visitors. This name is stored raw and later rendered via dangerouslySetInnerHTML in the Omnichannel Queue side panel (InquireSidePanelItem.tsx), injecting a real, clickable HTML link - pointing to any attacker-controlled domain, with arbitrary social-engineering text - into the DOM of any agent viewing the queue. | Mon 26 Oct |
| CVE-2026-44025 | 7.5 high | 0.00% | fluentd | Fluentd collects events from various data sources and writes them to files, RDBMS, NoSQL, IaaS, SaaS, Hadoop and so on. Prior to 1.19.3, Fluentd's Monitor Agent plugin in_monitor_agent exposes internal metrics and plugin information via a REST API, and responses from /api/plugins.json and related endpoints unintentionally include internal instance variables that may contain database passwords, API keys, or cloud credentials. This issue is fixed in version 1.19.3. | Sun 25 Oct |
| CVE-2026-55524 | 7.5 high | 0.00% | no CPE yet | PraisonAI is a multi-agent teams system. In versions prior to 1.6.58, the web_crawl tool performs its SSRF check only on the initially supplied URL, allowing the protection to be bypassed so the tool connects to attacker-chosen internal destinations. The check resolves the hostname once with socket.gethostbyname and rejects private/loopback/link-local results, but then passes the URL to a fetcher using httpx.Client(follow_redirects=True) (or urllib.request.urlopen when httpx is absent, which also follows redirects) that re-resolves the hostname at connect time with no further validation. This validate-here/fetch-there gap is exploitable through both HTTP redirects and DNS rebinding. If an attacker can influence URLs passed to web_crawl(), directly or through an agent/tool workflow, they can cause the PraisonAI host to fetch loopback, private-network, or cloud metadata endpoints reachable from that host, with the response body returned in the web_crawl() result. This issue has been fixed in version 1.6.58. | Sat 24 Oct |
| CVE-2026-55973 | 7.5 high | 0.00% | nlnetlabs, unbound | In NLnet Labs Unbound 1.23.0 up to and including 1.25.1, when 'dns-error-reporting: yes' is set, the EDNS Report-Channel option (code 18) from the last upstream response is read and uses the option's length as the length of the agent domain. When a domain name check is performed on the agent domain, the returned lenght is not used and if the agent domain is followed by garbage, those bytes are moved onto the tail of the synthetic '_er.' report query name. That query name is later used in the iterator via a subquery to send out the DNS Error Report and when Unbound tries to walk that query name during 'find_closest_of_type()', it strips labels using the query name length rather than stopping at the embedded root, walks one byte past it, and feeds the first garbage byte to 'dname_query_hash()' as a label length writing over the stack variable 'labuf'. One ordinary upstream response from a delegated zone the attacker controls is sufficient to terminate the daemon. | Sat 24 Oct |
| CVE-2026-56833 | 7.5 high | 0.00% | no CPE yet | PraisonAI is a multi-agent teams system. From version 3.8.1 to before version 4.6.59, PraisonAI's Dynamic Context module provides filesystem-backed history and terminal-log storage. The module also exports agent-callable tool factories. Those tools accept run_id and agent_id arguments from the tool caller. The underlying stores join those values into filesystem paths without rejecting absolute paths or .. traversal. Because run_id can be an absolute path and agent_id can contain traversal, a lower-trust prompt/user that can call these tools can read .jsonl and .log files outside the configured Dynamic Context base directory. This issue has been patched in version 4.6.59. | 23 Oct |
| CVE-2026-11810 | 7.5 high | 0.00% | no CPE yet | The UpdateHub firmware-update agent's probe handler (z_impl_updatehub_probe() in subsys/mgmt/updatehub/updatehub.c) parses the JSON metadata returned by the update server into a fixed two-level nested-array struct. After parsing it validates only the outer array length (objects_len != 2) and then dereferences objects[1].objects[0].objects.sha256sum via strlen() without checking that the inner object array of element [1] is non-empty. The metadata is attacker-influenceable network input: the agent fetches it over CoAP from the configured UpdateHub server during its routine OTA probe. A malicious or compromised update server (or, when DTLS is disabled, a network man-in-the-middle) can return a response whose second outer object array is empty. Because the parse target is zero-initialised, the corresponding objects[1].objects[0].objects.sha256sum pointer is NULL, and the subsequent strlen() dereferences address zero. The same defect exists in both the 'any boards' and 'some boards' metadata layouts. The resulting CPU fault is fatal under Zephyr's default error handling, halting or resetting the device, so the flaw is a remotely triggerable denial of service. Impact is limited to availability; it is a read from NULL with no out-of-bounds write, memory corruption, or information disclosure. The fix rejects metadata whose inner object array is empty before any dereference, on both layouts. | 23 Oct |
| CVE-2026-57146 | 7.5 high | 0.00% | no CPE yet | PraisonAI is a multi-agent teams system. From version 4.5.115 to before version 4.6.61, the published A2U advisory GHSA-f292-66h9-fpmf says unauthenticated A2U event streaming was fixed in praisonai 4.5.115. Current head still exposes the same A2U subscription and event routes without authentication when the operator starts the documented CLI entrypoint. The current CLI wrapper does not expose --api-key, does not install the common API-key middleware, and does not generate a token for A2U. It calls create_a2u_routes(app) directly. That helper only enforces auth if A2U_AUTH_TOKEN is already present; if the variable is missing, _authenticate_request() returns None and treats auth as disabled. This issue has been patched in version 4.6.61. | 23 Oct |
| CVE-2026-56834 | 7.5 high | 0.00% | no CPE yet | PraisonAI is a multi-agent teams system. From version 3.8.1 to before version 4.6.59, PraisonAI's Dynamic Context Discovery feature exposes artifact helper tools through ctx.get_tools(). The implemented artifact tools do not enforce that the supplied artifact_path is an artifact created by the configured store or that it lives under the configured artifact base directory. Instead, artifact_head, artifact_tail, artifact_grep, and artifact_chunk wrap the caller-supplied path directly into an ArtifactRef and then read it from the host filesystem. As a result, any prompt/user/tool-caller that can influence those tool arguments can read files readable by the PraisonAI process, such as project .env files, cloud credentials, SSH keys, source files, or other local data. This issue has been patched in version 4.6.59. | 23 Oct |
| CVE-2026-113619 | 7.5 high | 0.00% | no CPE yet | In Eclipse ioFog Controller versions 3.8.0 through 3.9.1, POST, PUT, and PATCH bodies on routes that enable template substitution are rendered by an uncapped Liquid engine before authentication and RBAC. The bearer check does not reject a missing or invalid token, and the auth rate limit applies only to user login routes. A small JSON body can stall or abort the Controller process. The anonymous caller does not receive the rendered result; the impact is availability of the control plane, including the agent API and in-process WebSocket relays on that process. | 20 Oct |
| CVE-2026-103366 | 7.5 high | 0.00% | no CPE yet | The Appointment Booking Plugin – LatePoint | Calendar & Scheduling for WordPress plugin for WordPress is vulnerable to Insecure Direct Object Reference in versions up to, and including, 5.7.3 via the `booking[id]` and `cart[id]` parameters on the `steps__reload_booking_form_summary_panel` and `steps__start` public routes (reachable through wp_ajax_nopriv_latepoint_route_call). This is due to `OsStepsHelper::set_required_objects()` feeding raw request parameters into `set_booking_object()` and `set_cart_object()`, which call `OsModel::set_data()` and load an arbitrary booking or cart row by primary key with no ownership check — the ownership guard the vendor added for the earlier disclosure lives only inside `OsStepsHelper::load_step()` and does not cover the sibling routes or `cart[id]` at all. This makes it possible for unauthenticated attackers to enumerate sequential booking / cart identifiers and read other customers' appointment details (service name, agent, date/time, duration, price breakdown, status) and other shoppers' pending checkout contents directly from the JSON responses. | 15 Oct |
| CVE-2025-65176 | 7.5 high | 0.00% | dynatrace, oneagent | An issue was discovered in Dynatrace OneAgent before 1.325.47. When attempting to access a remote network share from a machine where OneAgent is installed and receiving a "STATUS_LOGON_FAILURE" error, the agent will retrieve every user token on the machine and repeatedly attempt to access the network share while impersonating them. The exploitation of this vulnerability can allow an unprivileged attacker with access to the affected system to perform NTLM relay attacks. | 14 Oct |
| CVE-2026-107286 | 7.5 high | 0.00% | pydantic, pydantic_ai | Pydantic AI is a Python agent framework for building applications and workflows with Generative AI. From 2.10.0 until 2.53.0, streamed requests made through ConcurrencyLimitedModel or limit_model_concurrency can retain shared concurrency slots because anyio.CapacityLimiter associates an acquired slot with the borrowing task while streaming cleanup can run in a different task. Early stream termination, cancellation, consumer exceptions, or complete stream_text() consumption with debounce_by=0.1 can therefore leave capacity occupied, eventually preventing later requests that share the long-lived limiter from proceeding and causing a denial of service. Agent-level max_concurrency and non-streaming model requests are not affected. This issue is fixed in version 2.53.0. | 13 Oct |
| CVE-2025-11493 | 7.5 high | 0.00% | connectwise, automate | The ConnectWise Automate Agent does not fully verify the authenticity of files downloaded from the server, such as updates, dependencies, and integrations. This creates a risk where an on-path attacker could perform a man-in-the-middle attack and substitute malicious files for legitimate ones by impersonating a legitimate server. This risk is mitigated when HTTPS is enforced and is related to CVE-2025-11492. | 9 Oct |
| CVE-2026-82627 | 7.5 high | 0.00% | no CPE yet | The Uncanny Automator – AI + Automation for WordPress | AI Agent, AI Page Builder, Free AI Usage Included plugin for WordPress is vulnerable to PHP Object Injection in all versions up to, and including, 7.6.1.1 via deserialization of untrusted input. This makes it possible for authenticated attackers, with Subscriber-level access and above, to inject a PHP Object when a third-party integration plugin (such as PeepSo, MailPoet, WPForms, etc) is installed and a recipe is configured that stores attacker-controlled data as trigger meta. The additional presence of a POP chain within Uncanny Automator allows attackers to delete arbitrary files on the server. | 8 Oct |
| CVE-2026-35099 | 7.4 high | 0.00% | no CPE yet | Lakeside SysTrack Agent 11 before 11.5.0.15 has a race condition with resultant local privilege escalation to SYSTEM. The fixed versions are 11.2.1.28, 11.3.0.38, 11.4.0.24, and 11.5.0.15. | Mon 26 Oct |
| CVE-2026-33694 | 7.4 high | 0.00% | tenable, nessus, nessus_agent | This vulnerability allows an attacker to create a junction, enabling the deletion of arbitrary files with SYSTEM privileges. As a result, this condition potentially facilitates arbitrary code execution, whereby an attacker may exploit the vulnerability to execute malicious code with elevated SYSTEM privileges. | Mon 26 Oct |
| CVE-2026-45373 | 7.4 high | 0.00% | no CPE yet | CodeWhale is a DeepSeek + MiMo coding agent in terminal. Prior to 0.8.26, although SSRF is validated against hostnames that resolve to private IPv6 addresses, when providing the IPV6 in URL as http://[::1], the SSRF defenses do not work. This vulnerability is fixed in 0.8.26. | Mon 26 Oct |
| CVE-2026-40213 | 7.4 high | 0.00% | no CPE yet | OpenStack Cyborg before 16.0.1 uses rule:allow (check_str='@') as the default policy for multiple API endpoints. This unconditionally authorizes any request carrying a valid Keystone token regardless of roles, project membership, or scope. An authenticated user with zero role assignments can complete various actions such as reprogramming FPGA bitstreams on arbitrary compute nodes via agent RPC. | Mon 26 Oct |
| CVE-2026-45310 | 7.4 high | 0.00% | no CPE yet | CodeWhale is a DeepSeek + MiMo coding agent in terminal. Prior to 0.8.22, the fetch_url tool validates the initial URL's resolved IP address against a restricted-IP blocklist (is_restricted_ip()) to prevent SSRF attacks against internal services (cloud metadata endpoints, localhost, private networks). However, the HTTP client (reqwest) is configured to automatically follow up to 5 redirects (reqwest::redirect::Policy::limited(5)) without re-validating the redirect target against the same SSRF protections. This vulnerability is fixed in 0.8.22. | Sun 25 Oct |
| CVE-2026-61791 | 7.4 high | 0.00% | no CPE yet | OCO-Server is a self-hosted endpoint management application for computer inventory, software deployment, and mobile device management. Prior to 1.2.1, `Html::commandButton()` in `lib/Html.class.php` embeds an action URL containing a stored computer hostname into a JavaScript string in an inline `onclick` handler, using `htmlspecialchars(ENT_QUOTES)` only for HTML escaping. Browsers decode the HTML entities before interpreting the handler as JavaScript, allowing a crafted hostname to escape the string and execute script in an administrator's authenticated session when an affected command button is activated. `frontend/views/computer-details.php` passes `$computer->hostname` to this helper using `Computer::DEFAULT_COMPUTER_COMMANDS` in `lib/Models/Computer.class.php`, whose VNC, RDP, SSH, and Ping entries use `new_tab=false` and the vulnerable `window.location` branch. A remote attacker who possesses the shared `agent-registration-key` can use the `oco.agent.hello` handler in `frontend/api-agent.php` to store an unchecked hostname when `agent-self-registration-enabled` is enabled; self-registration is disabled by default. The vulnerable default command buttons require no custom command configuration. The injected script can abuse the administrator's privileges, including permission to deploy software jobs to managed computers. This issue is fixed in version 1.2.1. | 22 Oct |
| CVE-2026-100163 | 7.4 high | 0.00% | no CPE yet | The Appointment Booking Plugin – LatePoint | Calendar & Scheduling for WordPress plugin for WordPress is vulnerable to Authentication Bypass in versions up to, and including, 5.7.2 via the OsCustomerHelper::create_wp_user_for_customer() helper reached from the public process_step_customer() booking step. When the 'Use WordPress users as customers' option is enabled, the helper falls back from email_exists() to username_exists( $customer->email ), which matches any WordPress user whose user_login equals the victim customer's email even when that user's user_email is completely different, and links that WordPress user's ID into the victim's LatePoint customer row (only privileged roles are excluded; Subscriber is allowed). This makes it possible for authenticated attackers, with Subscriber-level access and above, to take over another customer's LatePoint account — accessing their bookings, service, agent, and calendar data, and having the victim customer's email overwritten with the attacker's on next login — by holding a Subscriber account whose user_login equals the victim's email and submitting an unauthenticated public booking step registration request. | 14 Oct |
| CVE-2026-107318 | 7.4 high | 0.00% | no CPE yet | @fastify/reply-from is a Fastify plugin that forwards requests to an upstream HTTP or HTTPS server. In versions prior to 12.7.0, all of the built-in HTTPS transports override the secure default and set rejectUnauthorized to false, so the proxy does not verify the TLS certificate of the upstream even when the application points it at an https upstream in the default configuration. An on-path network attacker can therefore impersonate the configured HTTPS upstream, read the credentials and request bodies the proxy forwards, and return forged responses that the application trusts. The issue is fixed in @fastify/reply-from 12.7.0, and users should upgrade to 12.7.0 or later. As a workaround, pass an explicit rejectUnauthorized true on the transport, supply an already configured undici instance, or use the undici global agent. | 9 Oct |
| CVE-2026-105477 | 7.3 high | 0.00% | no CPE yet | A vulnerability in Veeam Agent for Linux allows for local privilege escalation. Only installations that are managed by a Veeam Backup & Replication instance are affected. | Wed 28 Oct |
| CVE-2026-39306 | 7.3 high | 0.00% | praison, praisonai | PraisonAI is a multi-agent teams system. Prior to 1.5.113, PraisonAI's recipe registry pull flow extracts attacker-controlled .praison tar archives with tar.extractall() and does not validate archive member paths before extraction. A malicious publisher can upload a recipe bundle that contains ../ traversal entries and any user who later pulls that recipe will write files outside the output directory they selected. This is a path traversal / arbitrary file write vulnerability on the client side of the recipe registry workflow. It affects both the local registry pull path and the HTTP registry pull path. The checksum verification does not prevent exploitation because the malicious traversal payload is part of the signed bundle itself. This vulnerability is fixed in 1.5.113. | Mon 26 Oct |
| CVE-2026-40149 | 7.3 high | 0.00% | praison, praisonai | PraisonAI is a multi-agent teams system. Prior to 4.5.128, the gateway's /api/approval/allow-list endpoint permits unauthenticated modification of the tool approval allowlist when no auth_token is configured (the default). By adding dangerous tool names (e.g., shell_exec, file_write) to the allowlist, an attacker can cause the ExecApprovalManager to auto-approve all future agent invocations of those tools, bypassing the human-in-the-loop safety mechanism that the approval system is specifically designed to enforce. This vulnerability is fixed in 4.5.128. | Mon 26 Oct |
| CVE-2026-41288 | 7.3 high | 0.00% | watchguard, agent | Incorrect permission assignment for a resource in the patch management component of the WatchGuard Agent on Windows allows an authenticated local user to elevate their privileges to NT AUTHORITY\\SYSTEM. | Mon 26 Oct |
| CVE-2026-14893 | 7.3 high | 0.00% | no CPE yet | IBM Observability with Instana (Agent) Build 1.0.303 through 1.0.320 IBM Instana Node.js tracer component @instana/core version 6.2.1 is vulnerable to prototype pollution through its configuration normalization API. | 23 Oct |
| CVE-2026-115253 | 7.3 high | 0.00% | no CPE yet | An incorrect permission assignment vulnerability in the WatchGuard Agent's log directory allows a local, authenticated non-administrative user to create, modify, append to, or delete the agent's log files and write other files into the directory. This allows an attacker to tamper with or destroy the agent's security logs and stage further local attacks by placing executable content in a directory owned by a privileged service. | 23 Oct |
| CVE-2026-75857 | 7.3 high | 0.00% | no CPE yet | CodeWhale versions >= 0.8.41 and < 0.8.64 contain a vulnerability in the exec_shell_interact (alias exec_interact) tool, whose approval_requirement returns ApprovalRequirement::Auto. This overrides the default Required approval for code-executing tools, so LLM-controlled stdin is written into an already-approved long-running interactive shell (e.g., a python3 -i REPL, mysql, ssh, or sudo -i session) without any approval prompt. An attacker who can inject instructions via untrusted content the agent ingests (a fetched page, MCP result, or repo file) can cause commands to run at the privilege level of that approved process. Fixed in 0.8.64. | 22 Oct |
| CVE-2025-27234 | 7.3 high | 0.00% | no CPE yet | Zabbix Agent 2 smartctl plugin does not properly sanitize smart.disk.get parameters, allowing an attacker to inject unexpected arguments into the smartctl command. In Zabbix 5.0 this allows for remote code execution. | 19 Oct |
| CVE-2025-48963 | 7.3 high | 0.00% | no CPE yet | Local privilege escalation due to improper soft link handling. The following products are affected: Acronis Cyber Protect Cloud Agent (Linux, macOS, Windows) before build 40296. | 18 Oct |
| CVE-2025-59504 | 7.3 high | 0.00% | microsoft, azure_monitor_agent | Heap-based buffer overflow in Azure Monitor Agent allows an unauthorized attacker to execute code locally. | 18 Oct |
| CVE-2026-108593 | 7.3 high | 0.00% | no CPE yet | 9router 0.4.1 through 0.5.99 contains a configuration injection vulnerability in the POST /api/cli-tools/hermes-settings endpoint that allows authenticated dashboard users to inject arbitrary keys into the Hermes Agent config.yaml file. Attackers can submit a baseUrl containing double quotes and newlines to add hooks_auto_accept and a hooks.post_llm_call shell command, which Hermes Agent executes without approval after an LLM call. | 13 Oct |
| CVE-2026-40315 | 7.2 high | 0.00% | praison, praisonai | PraisonAI is a multi-agent teams system. Prior to 4.5.133, there is an SQL identifier injection vulnerability in SQLiteConversationStore where the table_prefix configuration value is directly concatenated into SQL queries via f-strings without any validation or sanitization. Since SQL identifiers cannot be safely parameterized, an attacker who controls the table_prefix value (e.g., through from_yaml or from_dict configuration input) can inject arbitrary SQL fragments that alter query structure. This enables unauthorized data access, such as reading internal SQLite tables like sqlite_master, and manipulation of query results through techniques like UNION-based injection. The vulnerability propagates from configuration input in config.py, through factory.py, to the SQL query construction in sqlite.py. Exploitation requires the ability to influence configuration input, and successful exploitation leads to internal schema disclosure and full query result tampering. This issue has been fixed in version 4.5.133. | Mon 26 Oct |
| CVE-2026-66138 | 7.2 high | 0.00% | no CPE yet | In OpenStack Ironic Python Agent through 11.6.0, a project-scoped user with the manager role can achieve arbitrary code execution on a running Ironic-Python-Agent via a maliciously constructed configuration, because the value of ntp_server is passed to a shell. | Sat 24 Oct |
| CVE-2026-57114 | 7.2 high | 0.00% | no CPE yet | PraisonAI is a multi-agent teams system. From version 4.5.126 to before version 4.6.59, PraisonAI's Async Jobs API validates webhook_url when a job request is parsed and again when the internal Job object is constructed. That validation blocks direct loopback/private targets, but it is not bound to the later network request. When a job completes, _send_webhook() passes the original hostname to httpx.AsyncClient.post() with no send-time validation, IP pinning, or guarded transport. An attacker-controlled hostname can therefore resolve to a public IP during Pydantic validation and later resolve to loopback/private/cloud-metadata infrastructure during webhook delivery. This bypasses the intended SSRF guard in current supported releases. This appears to be an incomplete fix / patch bypass for GHSA-8frj-8q3m-xhgm. This issue has been patched in version 4.6.59. | 23 Oct |
| CVE-2026-107858 | 7.2 high | 0.00% | no CPE yet | Missing Origin Validation in WebSockets (CWE-1385) and Permissive Cross-domain Policy with Untrusted Domains (CWE-942) in the development web server (ADK Web) in Google Agent Development Kit (ADK) for Java (com.google.adk:google-adk-dev versions 0.1.0 before 1.13.0 and com.google.adk:google-adk-maven-plugin versions 0.3.0 before 1.13.0) allow a remote attacker to control a locally running agent and read its data via a malicious web page visited by a developer while the server is running. The /run_live WebSocket endpoint accepts connections from any origin by default (hardcoded via setAllowedOrigins("*") in versions 0.1.0 through 1.5.0, and defaulting to * via adk.web.cors.origins in versions 1.6.0 through 1.12.0), and since version 0.4.0 the REST API returns Access-Control-Allow-Origin: * by default. Because the server has no authentication and requests originate from the developer's browser (so binding to localhost does not prevent exploitation), a cross-origin page can send messages to the agent (triggering registered tools), stream agent responses, and list, read, create, or delete sessions and artifacts. In versions 1.6.0 through 1.12.0, users can mitigate the issue by setting the adk.web.cors.origins property to an explicit allowlist of trusted origins (such as http://localhost:8000). The vulnerability is fixed in version 1.13.0, which enforces a same-origin default for CORS and the /run_live WebSocket endpoint, and rejects DNS-rebinding requests over loopback. | 23 Oct |
| CVE-2026-108853 | 7.2 high | 0.00% | no CPE yet | UnicomAI Wanwu before 0.6.3 contains an insecure direct object reference vulnerability that allows authenticated low-privileged users to delete other tenants' agent or RAG applications by supplying their appId. Attackers can send requests to DELETE /v1/appspace/app with guessed sequential assistant IDs to permanently delete victims' applications, workflows, conversations, and associated data. | 16 Oct |
Download these 806 as CSV · subscribe to this list.
EPSS scores: 2026-10-30. 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.