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-2026-105039 | 9.6 critical | 0.00% | no CPE yet | EnergizeLab station-middleware (NestJS) Docker images 1.6.5 through 1.8.2 do not check user roles on administrative APIs. Any authenticated low-privilege user can retrieve AI provider keys and change configuration and object-store contents. | 16 Oct |
| CVE-2025-61929 | 9.6 critical | 0.00% | cherry-ai, cherry_studio | Cherry Studio is a desktop client that supports for multiple LLM providers. Cherry Studio registers a custom protocol called `cherrystudio://`. When handling the MCP installation URL, it parses the base64-encoded configuration data and directly executes the command within it. In the files `src/main/services/ProtocolClient.ts` and `src/main/services/urlschema/mcp-install.ts`, when receiving a URL of the `cherrystudio://mcp` type, the `handleMcpProtocolUrl` function is called for processing. If an attacker crafts malicious content and posts it on a website or elsewhere (there are many exploitation methods, such as creating a malicious website with a button containing this malicious content), when the user clicks it, since the pop-up window contains normal content, the direct click is considered a scene action, and the malicious command is directly triggered, leading to the user being compromised. As of time of publication, no known patched versions exist. | 8 Oct |
| CVE-2026-103093 | 9.4 critical | 0.00% | servicenow, servicenow ai platform | ServiceNow has remediated a sandbox escape security issue that was identified in the ServiceNow AI Platform. This vulnerability could enable an authenticated user with the admin role to bypass Rhino sandbox restrictions and execute arbitrary code on the underlying host operating system, potentially enabling further unintended access. ServiceNow deployed an update to hosted instances, and ServiceNow provided the update to our partners and self-hosted customers. We are not currently aware of malicious exploitation against ServiceNow instances. We recommend customers promptly apply appropriate updates or upgrade to a patched release if they have not already done so. | Thu 29 Oct |
| CVE-2026-55413 | 9.4 critical | 0.00% | no CPE yet | ToolJet is the open-source foundation am AI-native platform for building and deploying internal tools, workflows and AI agents. Prior to 3.20.178-lts, any authenticated user with builder role (free tier) can overwrite a globally-shared marketplace plugin with arbitrary JavaScript that executes server-side with full Node.js access (require, process). The malicious code runs whenever any user on the instance triggers a query using that plugin — achieving both RCE and supply-chain compromise of the entire ToolJet deployment. This vulnerability is fixed in 3.20.178-lts. | Sun 25 Oct |
| CVE-2026-49776 | 9.3 critical | 0.00% | no CPE yet | Unauthenticated SQL Injection in GPTranslate – Multilingual AI Translation for WordPress: Automatically Translate Websites <= 2.32.6 versions. | Sun 25 Oct |
| CVE-2026-54030 | 9.3 critical | 0.00% | librechat | LibreChat is an enhanced ChatGPT clone that supports multiple AI providers. Prior to 0.8.5, LibreChat's MCP OAuth implementation does not validate that the resource parameter from OAuth Protected Resource metadata (RFC 9728) matches the configured MCP server URL, allowing a malicious MCP server to steal access tokens intended for a legitimate server. This vulnerability is fixed in 0.8.5. | Sun 25 Oct |
| CVE-2026-59515 | 9.3 critical | 0.00% | no CPE yet | Improper Neutralization of Special Elements used in an SQL Command ('SQL Injection') vulnerability in Sergey AIWU ai-copilot-content-generator allows Blind SQL Injection.This issue affects AIWU: from n/a through <= 1.5.4. | 24 Oct |
| CVE-2026-52842 | 9.3 critical | 0.00% | no CPE yet | Lightpanda is a headless browser designed for AI and automation. Prior to 0.3.1, Lightpanda searched for @ across the entire URL string instead of only the authority component when computing a page origin, so a URL such as `http://attacker.com/@victim.com/` was fetched from attacker.com but treated as `http://victim.com`, allowing a complete Same-Origin Policy bypass. This issue is fixed in version 0.3.1. | 24 Oct |
| CVE-2026-52843 | 9.3 critical | 0.00% | no CPE yet | Lightpanda is a headless browser designed for AI and automation. Prior to 0.2.9, Lightpanda fetch() and XMLHttpRequest unconditionally attached session cookies to every HTTP request, ignoring credentials: omit, credentials: same-origin, credentials: include, and XMLHttpRequest.withCredentials, allowing an attacker-controlled origin in a Lightpanda session to issue authenticated cross-origin requests against a victim origin. This issue is fixed in version 0.2.9. | 24 Oct |
| CVE-2026-50562 | 9.3 critical | 0.00% | no CPE yet | FastGPT is a knowledge-based AI application platform. At commit 22ebfacbb43311e9b73294040ae0eb87390c6bba and earlier, artifacts built from untrusted pull request code in .github/workflows/preview-docs-build.yml and .github/workflows/preview-fastgpt-build.yml can be downloaded by privileged workflow_run jobs in .github/workflows/preview-docs-push.yml and .github/workflows/preview-fastgpt-push.yml, allowing attacker-controlled Docker images from the document/ tree or FastGPT build context to be pushed to GHCR and, for documentation previews, deployed with secrets.KUBE_CONFIG_CN. | 24 Oct |
| CVE-2026-67426 | 9.3 critical | 0.00% | no CPE yet | Flyto2 Core is an execution kernel for automation and AI-agent workflows. Prior to 2.26.7, the standalone flyto-verification service in src/core/verification_service.py exposes unauthenticated POST /run on 0.0.0.0:8344 and uses client-supplied callback_url for an outbound POST with X-Internal-Key: $FLYTO_RUNNER_SECRET while bypassing target_allowed, allowing unauthenticated SSRF and runner secret exfiltration. This issue is fixed in version 2.26.7. | 24 Oct |
| CVE-2026-111693 | 9.3 critical | 0.00% | no CPE yet | DataBuff ai-apm-web through 0.1.10 contains a hard-coded default credentials vulnerability that allows unauthenticated attackers to gain administrator access using the built-in admin/Databuff@123 account. Attackers can POST these credentials to /webapi/api/v1/auth/login to obtain an administrator token and access all protected /webapi routes. | 23 Oct |
| CVE-2026-15643 | 9.2 critical | 0.00% | no CPE yet | AWS HealthLake MCP Server (awslabs.healthlake-mcp-server) is a Model Context Protocol server that enables AI assistants to interact with AWS HealthLake FHIR datastores. A server-side request forgery in the pagination handling component in AWS awslabs.healthlake-mcp-server before 0.0.14 on all platforms might allow a remote authenticated user to exfiltrate AWS temporary security credentials to an arbitrary endpoint via a crafted next_token parameter. The server does not validate that pagination URLs point back to the expected HealthLake endpoint, allowing an actor to redirect subsequent requests to an actor-controlled server. Its recommended to upgrade to version 0.0.14 or later. | 24 Oct |
| CVE-2026-103882 | 9.1 critical | — | acadlix, acadlix lms – ai quiz maker, online exam & course plugin for wordpress | The Acadlix LMS – AI Quiz Maker, Online Exam & Course Plugin for WordPress plugin for WordPress is vulnerable to arbitrary file deletion due to insufficient file path validation in the post_delete_assessment_file_by_id function in all versions up to, and including, 1.1.9. This makes it possible for authenticated attackers, with subscriber-level access and above, to delete arbitrary files on the server, which can easily lead to remote code execution when the right file is deleted (such as wp-config.php). The route's permission callback only verifies a generic wp_rest nonce that the plugin itself emits inside the public acadlixOptions JavaScript object on every frontend page, effectively reducing the authentication requirement to unauthenticated. | 14 h ago |
| CVE-2026-117574 | 9.1 critical | 0.00% | no CPE yet | A flaw was found in ai-gateway-payload-processing. An authenticated remote attacker can exploit this vulnerability by submitting a request specifying another tenant's model identifier in the request body or header. Because the service does not verify authorization before retrieving credentials from the shared store, it injects the target tenant's application programming interface (API) keys into the request. This flaw allows an attacker to perform unauthorized requests and consume another tenant's large language model (LLM) resources and quotas. | Wed 28 Oct |
| CVE-2026-48814 | 9.1 critical | 0.00% | no CPE yet | Network-AI is a TypeScript/Node.js multi-agent orchestrator. In versions 5.7.1 and earlier, the MCP SSE server allows unauthenticated cross-origin MCP tool invocation due to an empty default secret. This issue was partially addressed by CVE-2026-46701 in version 5.4.5 by closing the CORS flaw (with Access-Control-Allow-Origin now set only for localhost origins), but the empty-default-secret flaw described in the title remained: the SSE MCP server still defaulted to an empty secret, _isAuthorized() still returned true when the secret was empty, and a non-loopback bind only produced a warning. As a result, the server still ran fully unauthenticated by default. Any non-browser caller (for example, curl, SSRF, or a 0.0.0.0 bind) could invoke all 22 MCP tools (config_set, agent_spawn, blackboard_write, token_*) with no credentials. This issue was fixed in version 5.7.2. | Sun 25 Oct |
| CVE-2026-92702 | 9.1 critical | 0.00% | no CPE yet | Cocos AI is a confidential computing system for AI. In Cocos AI 0.8.2 and earlier, VerifyAttestationReportTLS in the AMD SEV-SNP verifier sets Policy.ReportData only when its caller supplies non-empty expected reportData. A relying-party integration that passes nil or empty expected reportData while relying on that value for current-session freshness can therefore accept SEV-SNP evidence unrelated to the current connection and make a trust, authorization, or application-data release decision for an unintended attestation context. This issue was fixed in Cocos AI 0.9.0 by removing the affected verifier path as part of the migration to post-handshake attestation. | 13 Oct |
| CVE-2026-92701 | 9.1 critical | 0.00% | no CPE yet | Cocos AI is a confidential computing system for AI. In Cocos AI 0.8.2 and earlier, VerifTeeAttestation in the Intel TDX aTLS verifier receives an expected current-session value but does not apply it to TdQuoteBodyPolicy.ReportData before validating TDX QuoteV4 evidence. The verifier can consequently accept structurally valid evidence containing mismatched or reused REPORT_DATA for the current TLS session, causing the relying party to release application data or make a trust or authorization decision for an unrelated attestation context. This issue was fixed in Cocos AI 0.9.0 by removing the affected verifier path as part of the migration to post-handshake attestation. | 13 Oct |
| CVE-2026-52705 | 9.0 critical | 0.00% | no CPE yet | Unauthenticated Arbitrary File Upload in SigmaForms Pro – AI Generated Forms <= 1.4.5 versions. | Mon 26 Oct |
| CVE-2026-48989 | 8.9 high | 0.00% | no CPE yet | Windows-MCP is an open-source project that integrates AI agents with Windows. In versions prior to 0.7.5, certain HTTP modes exposed the MCP control plane without authentication while enabling wildcard CORS (allow_origins=*, allow_methods=*, allow_headers=*). Because the same server also exposed a PowerShell tool that executes caller-controlled commands as the Windows user running Windows-MCP, attackers could reach the control plane from arbitrary origins or non-browser clients and achieve arbitrary PowerShell execution. This issue was fixed in version 0.7.5. | Mon 26 Oct |
| CVE-2026-63378 | 8.8 high | — | fctr-id, okta-ai-agent | TAKO AI Agent for Okta is a local-first AI agent that helps Okta administrators query tenant data using natural language. Prior to 3.0.1-beta, any authenticated user with a valid `fctr_session` cookie can submit arbitrary Python source in the `script_code` field of `POST /api/react/execute-script`, without administrator privilege. In `src/api/routers/react_stream.py`, the source is stored with `skip_discovery=True`, bypassing AI-based content filtering, and is passed to `validate_generated_code` in `src/utils/security_config.py`. `BLOCKED_PATTERNS` recognizes direct `__import__` calls but misses the import name passed as a string, `getattr` is in `ALLOWED_BUILTINS`, and the AST checker handles `ast.Name` and `ast.Attribute` targets but fails to handle `ast.Call` targets. Dynamic `getattr` call chains through `__builtins__` therefore evade validation. The `_write_temp_script` helper writes the accepted code to disk, and `asyncio.create_subprocess_exec()` executes it as the application process user, which is root in the default container. Command output is returned as Server-Sent Events through `GET /api/react/stream-react-updates?process_id=<id>`. The resulting container compromise can disclose environment secrets including Okta API tokens and the JWT key, expose the SQLite database, and permit backend-service pivoting and persistence. This issue is fixed in version 3.0.1-beta. | 24 h ago |
| CVE-2026-65566 | 8.8 high | — | meow apps, ai engine | Cross-Site Request Forgery (CSRF) vulnerability in Meow Apps AI Engine ai-engine allows Cross Site Request Forgery.This issue affects AI Engine: from n/a through 3.8.8. | Fri 30 Oct |
| CVE-2026-61546 | 8.8 high | 0.00% | no CPE yet | AnkiMCP is an Anki addon that implements an MCP server, enabling AI assistants to interact with the Anki flashcard application. Prior to version 0.21.0, `anki_mcp_server/mcp_server.py` globally disables FastMCP DNS-rebinding protection through `TransportSecuritySettings` with `enable_dns_rebinding_protection=False`, including for the default local Streamable HTTP listener. The defaults in `anki_mcp_server/config.py` enable `http_enabled` on `127.0.0.1:3141`, with an empty `http_path` selecting the root endpoint. No bearer token, API key, session secret, strict `Host` allowlist or strict `Origin` allowlist is enforced before `initialize`, `tools/list` and `tools/call` reach MCP dispatch. A victim who visits a malicious web page while Anki and the listener are running can have browser requests rebound to loopback, allowing the attacker to `initialize` MCP, enumerate tools, read the local collection, and perform enabled non-destructive note/deck creation or updates. The attacker needs neither LAN position nor a local process. Available operations depend on enabled tool filters; destructive tools are hidden unless `enabled_destructive_tools` is opted into, limiting deletion-level impact but not the exposed read/write operations. The reported harness confirms real MCP tool dispatch with substituted Anki bridge responses rather than demonstrating destructive changes to a real collection. Disabling protection for tunnels/proxies also disables it for the normal loopback configuration. This issue is fixed in version 0.21.0. | Wed 28 Oct |
| CVE-2026-40978 | 8.8 high | 0.00% | vmware, spring_ai | SQL injection vulnerability in Spring AI's `CosmosDBVectorStore` allows attackers to execute arbitrary SQL queries via crafted document IDs. Affected versions: Spring AI: 1.0.0 - 1.0.5 (fixed in 1.0.6), 1.1.0 - 1.1.4 (fixed in 1.1.5) | Mon 26 Oct |
| CVE-2026-40352 | 8.8 high | 0.00% | fastgpt | FastGPT is an AI Agent building platform. In versions prior to 4.14.9.5, the password change endpoint is vulnerable to NoSQL injection. An authenticated attacker can bypass the "old password" verification by injecting MongoDB query operators. This allows an attacker who has gained a low-privileged session to change the password of their account (or others if combined with ID manipulation) without knowing the current one, leading to full account takeover and persistence. This issue has been fixed in version 4.14.9.5. | Mon 26 Oct |
| CVE-2026-6859 | 8.8 high | 0.00% | redhat, instructlab, enterprise_linux_ai | A flaw was found in InstructLab. The `linux_train.py` script hardcodes `trust_remote_code=True` when loading models from HuggingFace. This allows a remote attacker to achieve arbitrary Python code execution by convincing a user to run `ilab train/download/generate` with a specially crafted malicious model from the HuggingFace Hub. This vulnerability can lead to complete system compromise. | Mon 26 Oct |
| CVE-2026-63944 | 8.8 high | 0.00% | linux, linux_kernel | In the Linux kernel, the following vulnerability has been resolved: Bluetooth: hci_sync: fix UAF in hci_le_create_cis_sync hci_le_create_cis_sync() dereferences conn->conn_timeout after releasing both rcu_read_lock() and hci_dev_lock(hdev). The conn pointer was obtained from an RCU-protected iteration over hdev->conn_hash.list and is not valid once these locks are dropped. A concurrent disconnect can free the hci_conn between the unlock and the dereference, causing a use-after-free read. The cancellation mechanism in hci_conn_del() cannot prevent this because hci_le_create_cis_pending() queues hci_create_cis_sync with data=NULL: hci_cmd_sync_queue(hdev, hci_create_cis_sync, NULL, NULL); While hci_conn_del() dequeues with data=conn: hci_cmd_sync_dequeue(hdev, NULL, conn, NULL); Since NULL != conn, the lookup in _hci_cmd_sync_lookup_entry() never matches, and the pending work item is not cancelled. Fix this by saving conn->conn_timeout into a local variable while the locks are still held, so the stale conn pointer is never dereferenced after unlock. This is the same class of bug as the one fixed by commit 035c25007c9e ("Bluetooth: hci_sync: Fix UAF on le_read_features_complete") which addressed the identical pattern in a different function. This vulnerability was identified using 0sec.ai, an open-source automated security auditing platform (https://github.com/0sec-labs). | Mon 26 Oct |
| CVE-2026-33760 | 8.8 high | 0.00% | langflow | Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to 1.9.0, Langflow's /api/v1/monitor router exposes 7 endpoints that perform read, write, and delete operations on user-owned resources — messages, sessions, build artifacts, and LLM transaction logs — without verifying that the authenticated requester owns the targeted resource. Any authenticated user can read, modify, rename, or permanently delete another user's data by supplying the target's resource ID or flow_id. This is a classic IDOR/BOLA vulnerability. Notably, the same source file (monitor.py) contains one correctly-implemented endpoint that uses an ownership check, demonstrating the correct pattern was known but inconsistently applied. This vulnerability is fixed in 1.9.0. | Mon 26 Oct |
| CVE-2026-54149 | 8.8 high | 0.00% | no CPE yet | MaxKB is an open-source AI assistant for enterprise. Prior to 2.10.0-lts, MaxKB tool import functionality in apps/tools/serializers/tool.py and MCP referencing mode in apps/application/chat_pipeline/step/chat_step/impl/base_chat_step.py do not consistently validate MCP transport type, allowing an authenticated user to import a .tool file containing stdio transport with malicious commands and trigger the configuration through an AI Chat node so MultiServerMCPClient executes arbitrary system commands. This issue is fixed in version 2.10.0-lts. | Sun 25 Oct |
| CVE-2026-61684 | 8.8 high | 0.00% | no CPE yet | FastGPT is a knowledge-based AI application platform. In 4.15.0-beta4, FastGPT plugin invoke reverse-call endpoints under /api/invoke/* authenticate only by verifying a JWT signed with INVOKE_TOKEN_SECRET, which defaults to the constant string token and was not set in official deployment templates. An unauthenticated attacker can self-sign an HS256 JWT and reach /api/invoke/userInfo to disclose cross-tenant user PII by attacker-supplied tmbId values, or /api/invoke/fileUpload to write attacker-controlled content into chat files. This issue is fixed in version 4.15.0-beta5. | 24 Oct |
| CVE-2026-32560 | 8.8 high | 0.00% | no CPE yet | Subscriber Local File Inclusion in MagicAI for WordPress - AI Text, Image, Chat, Code, and Voice Generator <= 1.4 versions. | 24 Oct |
| CVE-2026-64623 | 8.8 high | 0.00% | no CPE yet | Network-AI before 5.13.4 contains an improper cryptographic signature verification vulnerability in APSAdapter where the default local verifier accepts any non-empty string as valid. Unauthenticated attackers can submit forged APS delegation payloads with arbitrary scopes to bypass signature verification and obtain signed permission-grant tokens for sensitive resources including SHELL_EXEC. | 24 Oct |
| CVE-2026-16745 | 8.8 high | 0.00% | no CPE yet | A flaw was found in odh-dashboard, the web console component of Red Hat OpenShift AI (RHOAI). Due to incorrect network binding, a malicious actor within the cluster can bypass authentication and impersonate any user by providing an arbitrary access token. This allows an attacker to gain unauthorized access to the Kubernetes API, potentially leading to arbitrary code execution, privilege escalation, or information disclosure. | 24 Oct |
| CVE-2026-63831 | 8.8 high | 0.00% | linux, linux_kernel | In the Linux kernel, the following vulnerability has been resolved: mac802154: llsec: add skb_cow_data() before in-place crypto llsec_do_encrypt_unauth(), llsec_do_encrypt_auth(), llsec_do_decrypt_unauth(), and llsec_do_decrypt_auth() all perform in-place cryptographic transformations on skb data. They build a scatterlist with sg_init_one() pointing into the skb's linear data area and then pass the same scatterlist as both src and dst to the crypto API (e.g. crypto_skcipher_encrypt/decrypt, crypto_aead_encrypt/decrypt). On the RX path, __ieee802154_rx_handle_packet() clones the received skb before handing it to each subscriber via ieee802154_subif_frame(). The cloned skb shares the same underlying data buffer via reference counting. When llsec_do_decrypt() subsequently modifies this shared buffer in place, it corrupts data that other clones -- potentially belonging to other sockets or subsystems -- still reference. On the TX path, similar data sharing can occur when an skb's head has been cloned (skb_cloned() returns true). The fix is to call skb_cow_data() before performing any in-place crypto operation. skb_cow_data() ensures that the skb's data area is not shared: if the skb head is cloned or the data spans multiple fragments, it copies the data into a private buffer that can be safely modified in place. This is the same pattern used by: - ESP (net/ipv4/esp4.c, net/ipv6/esp6.c) - MACsec (drivers/net/macsec.c) - WireGuard (drivers/net/wireguard/receive.c) - TIPC (net/tipc/crypto.c) Without this guard, in-place crypto on shared skb data leads to: - Silent data corruption of other skb clones - Use-after-free when the crypto API scatterwalk writes through a page that has already been freed by another clone's kfree_skb() - Kernel crashes under concurrent 802.15.4 traffic with security enabled (KASAN/KMSAN reports slab-use-after-free) Found by 0sec (https://0sec.ai) using automated source analysis. | 22 Oct |
| CVE-2026-53281 | 8.8 high | 0.00% | linux, linux_kernel | In the Linux kernel, the following vulnerability has been resolved: iommu/vt-d: Avoid NULL pointer dereference or refcount corruption Commit 60f030f7418d ("iommu/vt-d: Avoid use of NULL after WARN_ON_ONCE") fixed a NULL pointer dereference in an unlikely situation partly. If dev_pasid is not found in the dev_pasids list, it remains NULL. However, the teardown operations are executed unconditionally, this lead to a NULL pointer dereference or refcount corruption. If the domain was never attached to this IOMMU, info will be NULL, which would cause an immediate dereference when checking --info->refcnt. Even if info is not NULL, decrementing the refcount without having removed a valid PASID might unbalance the count. This could lead to premature dropping of the refcount to 0, potentially causing a use-after-free for the remaining active devices sharing the domain. Fix it by returning early if dev_pasid is NULL, before executing the teardown operations. Issue found by AI review and suggested by Kevin Tian. https://sashiko.dev/#/patchset/20260421031347.1408890-1-zhenzhong.duan%40intel.com | 21 Oct |
| CVE-2026-90041 | 8.8 high | 0.00% | no CPE yet | In the Linux kernel, the following vulnerability has been resolved: HID: sony: clean up device list on probe failure sony_input_configured() adds some controllers to sony_device_list before HID core registers their input devices. input_register_device() can fail after the callback returns successfully. sony_probe() then observes that HID_CLAIMED_INPUT is clear and unwinds, but only stops the HID hardware. The devres-managed sony_sc is freed while its list node remains linked, so the next matching controller traverses freed memory. Initialize the list node and device ID to inactive states. Make list removal idempotent and run the driver-private cleanup on every probe failure path. This also makes a second cleanup safe when sony_input_configured() already unwound a partial initialization before sony_probe() handles the missing input claim. Found by 0sec (https://0sec.ai) using automated source analysis; verified against the HID input registration and probe unwind paths. | 21 Oct |
| CVE-2026-68432 | 8.8 high | 0.00% | no CPE yet | In the Linux kernel, the following vulnerability has been resolved: vxlan: require CAP_NET_ADMIN in the device netns for changelink A tunnel changelink() operates on at most two netns, dev_net(dev) and the sticky underlay netns vxlan->net. They differ once the device is created in or moved to a netns other than the one the request runs in. The rtnl changelink path checks CAP_NET_ADMIN only against dev_net(dev), so a caller privileged there but not in vxlan->net can rewrite a vxlan device whose underlay lives in vxlan->net. vxlan_changelink() validates and applies the new configuration against vxlan->net (vxlan_config_validate(vxlan->net, ...)) and can reopen the underlay socket in that netns, so the same reasoning as the tunnel changelink series applies here. Gate vxlan_changelink() with rtnl_dev_link_net_capable(), at the top of the op before any attribute is parsed, matching ipgre_changelink() and the rest of the "require CAP_NET_ADMIN in the device netns for changelink" series. Found by 0sec automated security-research tooling (https://0sec.ai). | 19 Oct |
| CVE-2026-68240 | 8.8 high | 0.00% | no CPE yet | In the Linux kernel, the following vulnerability has been resolved: drm/gpusvm: publish dpagemap early to avoid device mapping leak on error drm_gpusvm_get_pages() only stored the local dpagemap into svm_pages->dpagemap on the success path. If a later page failed (e.g. -EOPNOTSUPP when ctx->allow_mixed is false) and jumped to err_unmap, svm_pages->dpagemap was still NULL, so __drm_gpusvm_unmap_pages() skipped device_unmap() and leaked the device mappings already created. Assign svm_pages->dpagemap when the first device page is mapped so the err_unmap path can device_unmap() those mappings. This issue was found by Sashiko AI review. | 19 Oct |
| CVE-2026-68142 | 8.8 high | 0.00% | no CPE yet | In the Linux kernel, the following vulnerability has been resolved: geneve: require CAP_NET_ADMIN in the device netns for changelink A tunnel changelink() operates on at most two netns, dev_net(dev) and the sticky underlay netns geneve->net. They differ once the device is created in or moved to a netns other than the one the request runs in. The rtnl changelink path checks CAP_NET_ADMIN only against dev_net(dev), so a caller privileged there but not in geneve->net can rewrite a geneve device whose underlay lives in geneve->net. geneve_changelink() applies the new configuration against geneve->net: geneve_link_config() and the geneve_quiesce()/geneve_unquiesce() pair reopen the underlay sockets in that netns (geneve_sock_add() uses geneve->net), so the same reasoning as the tunnel changelink series applies here. Gate geneve_changelink() with rtnl_dev_link_net_capable(), at the top of the op before any attribute is parsed, matching ipgre_changelink() and the rest of the "require CAP_NET_ADMIN in the device netns for changelink" series. Found by 0sec automated security-research tooling (https://0sec.ai). | 19 Oct |
| CVE-2025-61592 | 8.8 high | 0.00% | anysphere, cursor | Cursor is a code editor built for programming with AI. In versions 1.7 and below, automatic loading of project-specific CLI configuration from the current working directory (<project>/.cursor/cli.json) could override certain global configurations in Cursor CLI. This allowed users running the CLI inside a malicious repository to be vulnerable to Remote Code Execution through a combination of permissive configuration (allowing shell commands) and prompt injection delivered via project-specific Rules (<project>/.cursor/rules/rule.mdc) or other mechanisms. The fix for this issue is currently available as a patch 2025.09.17-25b418f. As of October 3, 2025 there is no release version. | 19 Oct |
| CVE-2025-61593 | 8.8 high | 0.00% | anysphere, cursor | Cursor is a code editor built for programming with AI. In versions 1.7 and below, a vulnerability in the way Cursor CLI Agent protects its sensitive files (i.e. */.cursor/cli.json) allows attackers to modify the content of the files through prompt injection, thus achieving remote code execution. A prompt injection can lead to full RCE through modifying sensitive files on case-insensitive filesystems. This issue is fixed in a commit, 25b418f, but has yet to be released as of October 3, 2025. | 19 Oct |
| CVE-2025-64106 | 8.8 high | 0.00% | anysphere, cursor | Cursor is a code editor built for programming with AI. In versions 1.7.28 and below, an input validation flaw in Cursor's MCP server installation enables specially crafted deep-links to bypass the standard security warnings and conceal executed commands from users if they choose to accept the server. If an attacker is able to convince a victim to navigate to a malicious deeplink, the victim will not see the correct speedbump modal, and if they choose to accept, will execute commands specified by the attackers deeplink. | 19 Oct |
| CVE-2025-64107 | 8.8 high | 0.00% | anysphere, cursor | Cursor is a code editor built for programming with AI. In versions 1.7.52 and below, manipulating internal settings may lead to RCE. Cursor detects path manipulation via forward slashes (./.cursor/./././././mcp.json etc.), and requires human approval to complete the operation. However, the same kind of manipulation using backslashes was not correctly detected, allowing an attacker who had already achieved prompt injection or some other level of control to overwrite sensitive editor files without approval on Windows machines. This issue is fixed in version 2.0. | 19 Oct |
| CVE-2025-64108 | 8.8 high | 0.00% | anysphere, cursor | Cursor is a code editor built for programming with AI. In versions 1.7.44 and below, various NTFS path quirks allow a prompt injection attacker to circumvent sensitive file protections and overwrite files which Cursor requires human approval to overwrite. Modification of some of the protected files can lead to RCE. Must be chained with a prompt injection or malicious model attach. Only affects systems supporting NTFS. This issue is fixed in version 2.0. | 19 Oct |
| CVE-2025-9803 | 8.8 high | 0.00% | lunary | lunary-ai/lunary version 1.9.34 is vulnerable to an account takeover due to improper authentication in the Google OAuth integration. The application fails to verify the 'aud' (audience) field in the access token issued by Google, which is crucial for ensuring the token is intended for the application. This oversight allows attackers to use tokens issued to malicious applications to gain unauthorized access to user accounts. The issue is resolved in version 1.9.35. | 18 Oct |
| CVE-2026-105033 | 8.8 high | 0.00% | no CPE yet | The EnergizeLab AI Station (RK3562) firmware image ships with default root credentials and SSH enabled. An attacker with network access to the device can log in as root and take full control of it. | 16 Oct |
| CVE-2025-65530 | 8.8 high | 0.00% | cloudlinux, ai-bolit | An eval injection in the malware de-obfuscation routines of CloudLinux ai-bolit before v32.7.4 allows attackers to overwrite arbitrary files as root via scanning a crafted file. | 14 Oct |
| CVE-2026-34046 | 8.7 high | 0.00% | langflow, langflow-base | Langflow is a tool for building and deploying AI-powered agents and workflows. Prior to version 1.5.1, the `_read_flow` helper in `src/backend/base/langflow/api/v1/flows.py` branched on the `AUTO_LOGIN` setting to decide whether to filter by `user_id`. When `AUTO_LOGIN` was `False` (i.e., authentication was enabled), neither branch enforced an ownership check — the query returned any flow matching the given UUID regardless of who owned it. This allowed any authenticated user to read any other user's flow, including embedded plaintext API keys; modify the logic of another user's AI agents, and/or delete flows belonging to other users. The vulnerability was introduced by the conditional logic that was meant to accommodate public/example flows (those with `user_id = NULL`) under auto-login mode, but inadvertently left the authenticated path without an ownership filter. The fix in version 1.5.1 removes the `AUTO_LOGIN` conditional entirely and unconditionally scopes the query to the requesting user. | Mon 26 Oct |
| CVE-2026-110474 | 8.7 high | 0.00% | no CPE yet | This vulnerability exists in the CreditNirvana Agentic AI Call Centre application due to hardcoded static API Key in the client side JavaScript bundle. An unauthenticated remote attacker could exploit this vulnerability by downloading client side JavaScript files and extracting API key, which could lead to make arbitrary API calls to retrieve call transcripts and recordings through the application API of the targeted system. | Mon 26 Oct |
| CVE-2026-48716 | 8.7 high | 0.00% | no CPE yet | nanobot is a personal AI assistant. In versions 0.1.5.post3 and prior, the WhatsApp bridge in bridge/src/whatsapp.ts constructs a filesystem path using the fileName field from an incoming WhatsApp document message without sanitization. The WhatsApp bridge downloads media attachments and writes them to disk using a filename derived from the sender's message via documentMessage.fileName, which is concatenated with a prefix and its raw value is passed directly to path.join(mediaDir, outFilename). Node.js path.join resolves .. components, allowing an attacker to escape the intended media/ directory by sending a document with a crafted fileName such as ../../../.ssh/authorized_keys. Because the attacker also controls the file content (the downloaded buffer), this is a write-anywhere primitive — both path and content are attacker-controlled. A fix for this issue is planned for version 0.1.5.post4. | Sun 25 Oct |
Download these 743 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.