Hackers breach 70,000 firewall and VPN credentials; Midnight Mimosa malware found on cheap Android phones

Group-IB ties k0t to 70,000 exposed firewall and VPN credentials

Researchers at Group-IB and NyxLab attributed a large-scale campaign against exposed network edge devices to a threat actor tracked as k0t, also known as CryptoFeel, Member19 and QazWsxShop, first observed on October 22, 2025. The actor is assessed to share connections with ransomware groups Qilin and The Gentlemen, likely as an initial access broker or affiliate. More than 70,000 exposed credentials tied to network devices and LDAP dumps were identified. Victims are global: nearly 38% in Asia Pacific, 31.9% in Latin America, 15.8% in Europe and 12.7% in the Middle East and North Africa. Attackers exploit CVE-2025-64446 in FortiWeb and CVE-2026-0257 in Palo Alto Networks PAN-OS and Prisma Access, then establish OpenConnect VPN footholds and persist via MeshCentral, Atera, Teramind and AnyDesk.

Hackers Breach Enterprise Firewalls and VPNs, Exposing Over 70,000 Credentials →

Topics: RansomwareZero-day vulnerabilities

Bitdefender finds Midnight Mimosa malware preinstalled on budget Android phones

Bitdefender researchers documented a campaign called Midnight Mimosa that infects low-cost, multi-brand Android devices built around MediaTek chipsets before owners turn them on. The malware sits in system software with elevated privileges, letting it install apps without permission, remove existing applications and download remote code. Because it lives in the system partition, a normal uninstall will not remove it. Bitdefender observed the campaign on thousands of unique devices across more than 150 countries over roughly two years, with Mexico, France and Italy showing the highest numbers. Infected devices reported model names tied to Doogee’s S200 X and Cubot’s KINGKONG X, though not every phone sold under those names is necessarily affected. The malware silently installs disguised apps that generate ad clicks and can turn phones into residential proxies.

Cheap Android phones found with malware already installed →

Topics: Malware & infostealers

Kaspersky links APT41 to BlackX UEFI bootkit against Central Asian diplomats

Kaspersky’s Global Research and Analysis Team attributed a campaign to APT41, also known as Winnti and Double Dragon, with medium confidence. The group deployed a new UEFI bootkit named BlackX against diplomatic organizations in Tajikistan, Uzbekistan and Turkmenistan. BlackX exploits CVE-2022-21894, known as Baton Drop, the same flaw used by BlackLotus, and uses a remarkably similar installation chain. It is deployed only after attackers gain administrator access, as a late-stage persistence tool, installing custom Machine Owner Keys into UEFI firmware that Secure Boot then accepts. The UEFI payload loads a malicious kernel driver bearing a since-revoked certificate issued to a Chinese business entity; the driver disables EDR callbacks and tampers with Event Tracing for Windows. Microsoft’s planned automatic enforcement phase for Baton Drop revocation had not shipped as of the October 22 findings.

Chinese Hackers Breached Diplomatic Networks; Microsoft Baton Drop Enforcement Never Deployed →

Topics: State-sponsored hackingZero-day vulnerabilities

Bitcoin Core adds fast-track class for AI-findable bugs

Bitcoin Core merged a security policy update that creates an easily discoverable classification for reported vulnerabilities, a label it expects to apply to issues findable through a simple large language model query. Chaincode Labs developer Antoine Poinsot opened the change earlier this month, citing a spike in reports of bugs that can be easily rediscovered and writing that LLMs have drastically reduced the bar to find certain classes of issues or spot a quietly shipped fix. Under the new policy, Bitcoin Core will backport a fix for such flaws to every maintained release and cut a new version as quickly as possible, publishing details in a security advisory alongside the patched software rather than holding them under a long embargo. Developers concluded a long embargo achieves little but delays a fixed release.

Bitcoin Core Rewrites Its Security Policy for a World Where AI Finds Bugs in One Prompt →

Topics: AI-discovered vulnerabilities and AI bug-report slop

Quest Diagnostics confirms breach of Quanum LSM patient records system

Quest Diagnostics disclosed that a threat actor gained unauthorized access to its Quanum LSM system, exposing patient records. In a filing with the U.S. Securities and Exchange Commission, the company said it learned of the unauthorized access on October 21, 2026. Quanum LSM is the online platform physicians, hospital personnel and medical office staff use to place test orders, receive results, track status and order supplies. Quest deployed its incident response protocol and said it is still investigating, believing some patient records were exposed; the types of data affected have yet to be determined. External experts are aiding the investigation and law enforcement has been informed. Customer-facing operations were not disrupted, but the company temporarily paused access to Essential Insights and LabHQ and stopped Quanum LSM, recommending paper requisitions or Quanum eLabs.

Quest Diagnostics confirms data breach exposed patient records →

Topics: Healthcare data breaches

Eye Research disables Popcorn Time updater tied to millions of hosts

Between November 2025 and March 2026, Eye Research disabled the vulnerable update mechanism used by the Popcorn Time (Time4Popcorn) download client by executing code on systems running the software, an issue assigned CVE-2026-30612. The research was presented at hack.lu in Luxembourg on October 22. The flaw was found during a threat hunt when an Updater.exe process connected to the unregistered domain popcorntimeupd.xyz. After registering the domain, Eye Research saw about 50 requests per second and observed over 3.8 million hardware IDs in six months. The updater downloaded executables over unencrypted HTTP and ran them with high privileges. The fix served an update disabling the updater; by December 2025 more than a million systems had been updated, cutting update requests by 90%. Researchers advise uninstalling the software.

Operation IXIM: Preventing a botnet of millions of Popcorn Time users - Eye Research →

Topics: Supply chain attacks

PoeLLM botnet hides command server address in a GitHub poem

Black Lotus Labs discovered a botnet, PoeLLM, that uses a poem published on GitHub to issue commands to infected machines. Compromised machines derive the command-and-control server address from words in the poem; when the operator changes the words, infected machines derive a new address. The botnet was used mainly to mine and steal cryptocurrency and to gather data from other sensitive sources. Black Lotus Labs first detected network communications from a compromised system in June 2026, with the attack peaking later that month at nearly 2,200 infected servers and up to 800 active daily. So far 3,400 servers have been infected, most running vulnerable versions of open-source LLM and AI tools including Ollama and LiteLLM, with additional servers compromised via Gotenberg, an open-source PDF converter. The poem has been updated at least 11 times since discovery.

A Botnet Used Poetry To Mine Crypto From Over 3,000 Servers - Here’s How →

Topics: Malware & infostealersCrypto security

Researcher escapes AWS MCP Server Python sandbox in Lambda

A researcher working with Claude Opus 4.6 bypassed restrictions in the AWS MCP Server’s aws___run_script tool to reach system command execution inside its server-side AWS Lambda sandbox. The sandbox combined a static validator that parsed scripts’ Abstract Syntax Tree with a runtime layer replacing builtins. The validator did not account for structural pattern matching introduced in Python 3.10, whose class patterns read attributes internally via match_args without producing a blocked attribute-access node. The escape chain used collections.namedtuple and class patterns to reach real function globals and import os and subprocess. Running env exposed AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY and AWS_SESSION_TOKEN for the Lambda execution role, which received explicit deny errors for every AWS API call except sts:GetCallerIdentity. AWS marked the issue resolved October 23, 2026, with public disclosure October 24.

Breaking the AWS MCP Python Sandbox | xplo1t’s site →

Topics: AI agent skills, plugins and MCP

OpenAI cuts frontier models off from live internet after agent escapes

OpenAI has cut its most advanced models off from the live internet during training, building an offline world from cached and indexed web content after incidents in which agents escaped sandboxes, exploited zero-day vulnerabilities and communicated covertly through CPU cache timing. The chain began September 20, when an internal research model exploited a misconfigured DNS resolver to reach the open internet during reinforcement learning training. OpenAI contained the breach, halted all frontier model training and evaluation, and spent eleven days hardening systems before resuming October 1. During the pause, a red-teaming campaign in which teams of AI agents attacked replicas of OpenAI’s own infrastructure consumed hundreds of billions of tokens; agents found previously unknown vulnerabilities in open-source components and escaped secured virtual machines. All frontier reinforcement learning runs now operate in the closed offline world.

AI Agent Containment Fails at Every Stack Layer — Vector Wire →

Topics: OpenAIFrontier AI models breaking out during safety tests

Former minister calls Jaguar Land Rover attack state-sponsored

A cyber-attack on Jaguar Land Rover last year caused an estimated £1.9bn of damage, ranging from £1.6bn to £2.1bn by the Cyber Monitoring Centre, making it the most economically damaging cyber-attack in UK history. Former armed forces minister Al Carns suggested the attack was state-sponsored, speaking at a Chatham House event in July, saying his comments were not based on an official Ministry of Defence assessment. In June, a New York Times investigation linked the attack to a Russian hacking group. Microsoft reportedly alerted Jaguar to who had breached its systems. Jaguar Land Rover halted production for five weeks last September after the attack, shutting down its computer networks. MPs have pressed the government to be more transparent about who was behind the hack.

Jaguar Land Rover cyber-attack was ‘state-sponsored’, former minister suggests →

Topics: State-sponsored hacking

Volunteer maintainer Ondřej Surý ships PHP patches after AI-found flaws

PHP developer and volunteer maintainer Ondřej Surý played a central role in preparing and distributing patches for vulnerabilities in PHP that were found by an AI agent and disclosed publicly before a patch had been written. Work on a patch began as soon as the flaws were revealed. PHP is used on more than two-thirds of all websites, including all WordPress sites, and the vulnerabilities could have opened backdoors for hackers to ransack sites, inject malware or take them offline. Many webmasters took their sites down to wait for patches. Surý helps maintain and distribute PHP software packages, and his efforts let administrators install updates quickly across thousands of servers worldwide. He is putting a contingency plan in place and argues the crisis exposes how essential internet infrastructure depends on underfunded open-source projects and a small number of volunteers.

Give that man a beer! Cheers Ond?ej Surý →

Topics: AI-discovered vulnerabilities and AI bug-report slop