threatdeck
Live 14:45 UTC ·

All decks

ThreatDeck — 2026-11-08

12 cards today · 3 actively exploited or must-patch · 6 affecting AI tooling.

  1. 01

    Act Now Critical KEV CVE-2026-119977 AI-assisted

    CISA adds WordPress Core TOCTOU race condition (CVE-2026-119977) to KEV

    CISA added CVE-2026-119977 to its Known Exploited Vulnerabilities catalog on 2026-11-03. The WordPress Core flaw is a time-of-check time-of-use race condition that could let unauthenticated attackers execute code on the server during the upgrade window. It stems from predictable, web-accessible staging paths used during full-package core upgrades. The federal due date was 2026-11-06, and use in ransomware campaigns is listed as unknown.

    Why it matters. Any internet-facing WordPress site performing a full-package core upgrade may be exposed to unauthenticated code execution, which could give an attacker control of the web server. CISA's KEV listing indicates the flaw is exploited in the wild.

    Do this. Apply the vendor's mitigations or fixed WordPress Core release per vendor instructions; the federal due date of 2026-11-06 has passed.

    2026-11-03 · score 61 · CISA Known Exploited Vulnerabilities

  2. 02

    Act Now High KEV CVE-2026-19868 AI-assisted

    CISA adds SonicWall SonicOS SSLVPN SAML/MFA bypass (CVE-2026-19868) to KEV

    CISA added CVE-2026-19868 to its Known Exploited Vulnerabilities catalog on 2026-11-06. The flaw in SonicWall SonicOS SSLVPN is a use of single-factor authentication issue. A user with valid credentials for an alternate authentication method may bypass SAML authentication controls, including SAML-enforced MFA. The federal due date is 2026-11-20, and use in ransomware campaigns is listed as unknown.

    Why it matters. Organizations that rely on SAML-enforced MFA for SonicWall SSLVPN access may not have that protection in practice. An attacker holding valid credentials for an alternate authentication method could gain VPN access without the MFA step. Exploitation is confirmed by CISA's KEV listing.

    Do this. Apply SonicWall's vendor-provided mitigations for CVE-2026-19868 to all SonicOS SSLVPN devices as soon as possible.

    2026-11-06 · score 54 · CISA Known Exploited Vulnerabilities

  3. 03

    Act Now High KEV CVE-2024-27834 AI-assisted

    CISA adds Apple CVE-2024-27834 pointer authentication bypass to KEV catalog

    CISA added CVE-2024-27834 to its Known Exploited Vulnerabilities catalog on 2026-11-04. The flaw is an improper authentication issue in Apple iOS, iPadOS, macOS, Safari, tvOS and watchOS. An attacker with arbitrary read and write capability could bypass pointer authentication. The federal due date is 2026-11-18. Known ransomware campaign use is listed as unknown.

    Why it matters. Inclusion in the KEV catalog indicates the flaw is being exploited. An attacker who already has memory read/write capability can bypass pointer authentication, a defense against code-reuse attacks. This weakens exploit mitigations across Apple devices and browsers.

    Do this. Apply Apple's current security updates for iOS, iPadOS, macOS, Safari, tvOS and watchOS across all managed devices.

    2026-11-04 · score 54 · CISA Known Exploited Vulnerabilities

  4. 04

    AI Stack Alert Critical CVE-2026-62863 AI-assisted

    MoLing MCP server before 0.5.0: unauthenticated SSE endpoints allow browser-based tool calls

    In MoLing versions before 0.5.0, SSE mode (started with `-l`) exposes `/sse` and `/message` without authentication. The `mcp-go` dependency sets `Access-Control-Allow-Origin: *`. A malicious web page can read the `sessionId` from the `/sse` endpoint event, including on a loopback listener, and then send MCP tool calls to `/message`. The issue is fixed in 0.5.0 and has a CVSS base score of 9.6.

    Why it matters. Anyone running MoLing in SSE mode can be attacked by simply visiting a malicious web page. Depending on the configured tools, an attacker can read or write local files, execute commands, or take other actions on the victim's machine.

    Do this. Upgrade MoLing to version 0.5.0 or later.

    2026-10-30 · score 45 · NVD (AI keyword search)

  5. 05

    AI Stack Alert Critical CVE-2026-116692 AI-assisted

    MLflow 3.12.0–3.16.x: unauthenticated RCE via third-party scorer deserialization (CVE-2026-116692)

    MLflow 3.12.0 added third-party scorer deserialization. The server imports a request-controlled module path under four allowed package prefixes and instantiates a request-controlled class. A caller can create an experiment whose artifact location sits inside an allowed package directory, upload a Python module as a run artifact, and then trigger its import through the scorer invoke endpoint. Affected versions are mlflow >= 3.12.0rc0 and < 3.17.0, and the fix is in 3.17.0.

    Why it matters. On default no-auth MLflow servers, a remote caller can run arbitrary Python code with the server process's privileges. The official image runs as uid 0, so an attacker could take over the host or container and reach model, experiment and artifact data. With basic auth enabled, the scorer endpoint does no resource-level permission check, so any user who can create experiments and upload artifacts may still reach it.

    Do this. Upgrade mlflow to 3.17.0 or later on all tracking servers.

    2026-10-27 · score 45 · GitHub Security Advisories (AI packages)

  6. 06

    AI Stack Alert Critical CVE-2026-120338 AI-assisted

    Prefab UI 0.14.0–0.20.2 Generative UI sandbox escape via Deno; fixed in 0.21.0

    In prefab-ui 0.14.0 through 0.20.2, the Generative UI sandbox runs Python in Pyodide inside a Deno subprocess started with unrestricted read, write and net permissions. The Deno API is left on the JavaScript global object, so supplied Python can read and write files as the server OS user. The usual remote path is FastMCP's GenerativeUI provider and its generate_prefab_ui tool. Version 0.21.0 revokes these permissions before accepting code.

    Why it matters. An attacker who can supply code to the sandbox can read and write files as the server user. They can also persist code execution by writing a .pth file into the virtualenv. FastMCP servers with Generative UI enabled are exposed to any client without authentication, or to any client allowed to call tools if authentication is on. The CVSS base score is 10.

    Do this. Upgrade prefab-ui to 0.21.0 or later.

    2026-11-01 · score 44 · NVD (AI keyword search)

  7. 07

    AI Stack Alert Critical CVE-2026-119978 AI-assisted

    Flowise through 3.1.4: vm2 sandbox escape to OS command execution (CVE-2026-119978)

    Flowise through 3.1.4 has a sandbox escape that lets authenticated users run OS commands. It abuses allowlisted langchain and mammoth modules when ALLOW_BUILTIN_DEP=true. JavaScript submitted to /api/v1/node-custom-function writes a file into the mammoth package via LocalFileStore and requires it, escaping vm2 as the Flowise process user. The CVSS base score is listed as 9.

    Why it matters. Any authenticated Flowise user on an affected instance with ALLOW_BUILTIN_DEP=true can run commands with the privileges of the Flowise process. This can expose host data, API keys and connected systems.

    Do this. Check whether Flowise instances run version 3.1.4 or earlier and check the vendor for a fixed release.

    2026-10-30 · score 41 · NVD (AI keyword search)

  8. 08

    AI Stack Alert Critical CVE-2026-119050 AI-assisted

    Agno PandasTools unsafe reflection (CVE-2026-119050) allows code execution via prompt injection

    Agno through version 3.1.11 has an unsafe reflection flaw in PandasTools. Model-controlled function names are passed to getattr(), so arbitrary pandas functions can be invoked. Via prompt injection, an attacker can make the agent call pandas.read_pickle on attacker-controlled paths for code execution, or call other readers to read local files. The CVSS base score is 9.2.

    Why it matters. Organizations running Agno agents with PandasTools that process untrusted input are exposed. An attacker who can influence the model's input may gain code execution on the host or read local files.

    Do this. Check whether any Agno deployments use PandasTools and are on version 3.1.11 or earlier.

    2026-10-29 · score 41 · NVD (AI keyword search)

  9. 09

    AI Stack Alert Critical CVE-2026-117574 AI-assisted

    CVE-2026-117574: ai-gateway-payload-processing leaks cross-tenant LLM API keys (CVSS 9.1)

    A flaw in ai-gateway-payload-processing lets an authenticated remote attacker specify another tenant's model identifier in the request body or header. The service does not check authorization before retrieving credentials from the shared store. It then injects the target tenant's API keys into the request. The CVSS base score is 9.1.

    Why it matters. In multi-tenant AI gateway deployments, any authenticated user could make requests using another tenant's LLM credentials. The attacker gains unauthorized use of that tenant's LLM resources and quotas, which can cause cost and service impact.

    Do this. Identify whether ai-gateway-payload-processing is deployed in your AI gateway stack, and check the vendor or project for a fixed version.

    2026-10-28 · score 41 · NVD (AI keyword search)

  10. 10

    New Technique High TD-2026-0019 AI-assisted

    3AM ransomware affiliate used hybrid human-AI operation against UK school estate

    Bitdefender reports that a 3AM ransomware affiliate compromised a UK school estate between September 10 and 19, 2026, encrypting 37 machines. Parts of the post-compromise activity were run by a human operator and parts behaved like an AI agent, based on command timing patterns. The researchers could not determine the initial access method.

    Why it matters. Organizations that hold privileged credentials for multiple domains or store backup credentials in shared locations are exposed to fast lateral movement. The hybrid model lowers the skill needed for attackers to run repetitive configuration work at scale.

    Do this. Keep administrative accounts for separate domains separate, and do not store privileged credentials for several domains in one place.

    2026-11-04 · score 33 · Infosecurity Magazine

  11. 11

    New Technique High TD-2026-0039 AI-assisted

    Cursor security: 10 documented attacks, and Workspace Trust is still off by default

    Adversa AI documents ten attacks on Cursor from August 2025 to October 2026. The three entries with CVEs have no confirmed in-the-wild exploitation, and the four that caused confirmed harm have no CVE. Cursor's documentation says Workspace Trust is disabled by default.

    Why it matters. Standard CVE-based scanning would miss most of the confirmed harm to Cursor users. Developers who open untrusted repositories, or install unvetted extensions and MCP servers, can run attacker-controlled code with the editor's full permissions.

    Do this. Enable Workspace Trust for developer Cursor installs, and do not open unknown repositories until they are reviewed.

    2026-11-03 · score 33 · Adversa AI

  12. 12

    Signal High AI-assisted

    Fake OpenAI, SpaceX, Anthropic event lures steal Microsoft and Google logins

    Attackers use fake OpenAI, SpaceX, and Anthropic event registration pages to phish corporate Microsoft and Google accounts. The pages link two phishing-as-a-service platforms: Kali365 for Microsoft device code phishing and VoidProxy as an AiTM reverse proxy for Google sign-ins and as a second stage for Microsoft. Push Security reported detecting and blocking visits in late October 2026.

    Why it matters. Employees who sign in with corporate accounts on these pages can hand over OAuth tokens, session cookies, and MFA responses, letting attackers reach Microsoft 365, Google Workspace, and connected SSO apps without the password.

    Do this. Block or restrict OAuth 2.0 device code authentication for users who do not need it, using Microsoft Entra conditional access.

    2026-11-05 · score 32 · BleepingComputer

Following

Follow the vendors, products, AI tools, techniques and actors you care about. Matching stories, CVEs and exploited bugs are marked, a Following tab appears in the feed, and returning visits show what changed. Saved only in this browser.

Presets

Get updates

A personal RSS feed of new stories on everything you follow. Paste it into any feed reader, Slack or Teams RSS app.