Google takes down a 1.3M-device botnet

Hi there, this is your daily ☕️ Cyberpresso.

Google kills 1.3M-device Android botnet LINK

• Google led a takedown that cut off the servers behind a botnet of more than 1.3 million Android devices, mostly TV set-top boxes, which were used for ad fraud and routing strangers' internet traffic.
• Google says Shenzhen maker Videostrong intentionally hid a firmware backdoor called x.cpe that turned off Play Protect, installed apps silently, captured screens and opened remote shells across 206 models in countries led by Malaysia and Brazil.
• Because the backdoor runs as a signed system app, it reinstalls on reboot, so experts tell owners to throw the boxes out; Google banned Videostrong from building certified Android devices and is tightening firmware checks.

🇧🇷 Brazil crypto exchange hacked for $46M LINK

• Mercado Bitcoin, a Brazilian crypto exchange, stopped withdrawals and deposits today after hackers stole an estimated $46 million in crypto, calling the pause a precautionary move while it investigates alongside authorities.
• The exchange says customers were unharmed and no data was breached, and it has shared the wallet addresses where stolen funds went so analysts can trace the money as thieves sell off memecoins and altcoins.
• The $46 million loss is small next to last week's Binance hack of over $6 billion, which the US Secret Service is investigating and is believed tied to the North Korea-linked group Trader Traitor.

🇨🇳 Chinese hackers used AI in cyberattacks LINK

• Chinese hackers known as Vapor Panda ran an AI tool called ARTEX against more than 200 organizations in 14 countries, hitting government, telecom, pharmaceutical, semiconductor, defense, nuclear, and financial targets since October 17.
• CrowdStrike says the AI agents did reconnaissance, exploited flaws, deployed web shells, and stole credentials, with human operators giving orders in Chinese and telling the agents to skip machines protected by its Falcon endpoint software.
• The attackers used DeepSeek and a second commercial language model to run several break-ins at once, compromising at least nine organizations and burning over 1.4 billion input tokens in one intrusion alone.

Industrial software flaw spreads Linux malware LINK

• Attackers are abusing a top-severity flaw in Node-RED, a workflow tool used in industrial control systems, to plant a Linux bot by sending a malicious automation flow that runs shell commands automatically once deployed.
• Tracked as CVE-2026-58115 and rated a perfect 10, the bug lets unauthenticated users reach programming nodes on affected Siemens SIMATIC IoT2050 devices; VulnCheck saw active exploitation from a single Brazil-linked IP starting on Friday.
• The scan-and-infect activity hit the Node-RED flaw 16 times and pummeled eight separate CVEs, suggesting opportunistic mass scanning; VulnCheck urges checking internet exposure and applying updates, and added it to its own exploited-bug list.

Magento flaw lets hackers hijack servers LINK

• Attackers are exploiting a Magento and Adobe Commerce flaw called StyleSmuggler (CVE-2026-75650) to run code on online stores, dropping a shell script, a Go remote access trojan called PSK Agent, and a PHP webshell for lasting control.
• The flaw is actively exploited across Europe, North America, and Asia-Pacific through the /paypal/transparent/response/ endpoint, and the dropper script had a 0% detection rate on VirusTotal, letting it slip past file-reputation checks entirely.
• Adobe's VULN-39341 hotfix closes the entry point but won't remove malware planted before patching, so admins should hunt for the dropper, webshell, and fake process names, then rotate all exposed Magento credentials and keys.

Strategies & Tactics

> Beyond DKIM key rotation: understanding email replay attacks: Rotating a DKIM email signing key doesn't stop replay attacks because the old key stays valid; deleting it from DNS closes that window.

> AI coding agents are installing MCP servers that will never be patched: AI coding agents install outdated, unpatchable software packages they remember from training, exposing machines to known security holes that monitoring tools miss entirely.

> Anthropic Enables EDR Security Monitoring for Claude Cowork AI Agents on Windows and macOS: Anthropic lets security teams install monitoring software inside Claude Cowork's local virtual machine, exposing AI agent activity that normal computer monitoring tools cannot see.

> Discovering a Backdoor and Flaws in More Encrypted USB Drives: AI-assisted reverse engineering broke several "encrypted" USB drives in minutes, exposing fixed-password backdoors and hardcoded keys that decrypt files regardless of the user's password.

> SPECTER attack leaks air-gapped data through USB power controls: Researchers showed that already-compromised air-gapped computers can leak passwords or keys by toggling USB power controls, detectable via a nearby phone's charging status.

> Hackers Could Exploit iPhone BootROM Flaws to Compromise Cryptocurrency Wallet Security: Storing crypto recovery phrases on old offline iPhones is risky because unfixable startup-code flaws let anyone with physical access run malicious code.

Other news you might like

• BITTER APT Using Fake Diplomatic Documents to Deliver PapRAT Spyware LINK

• Russia-Aligned Gorbag Hackers Exploit WinRAR Flaw to Deploy Data-Stealing Malware LINK

• DarkBlinders Compromises Israeli Target and Iraqi Kurdish Government Cloud Environment LINK

• Apple warns users in 99 countries of mercenary spyware LINK

• Hackers Hijacked Linux Hosting Server Security Tool cPGuard and Used It as a Backdoor LINK

• Cloud-hosted AI workflow tool has environment variables snatched in security incident LINK

• Hackers Use ARTEX AI to Breach WorkOS and Access Customer Slack Conversations LINK

Trending tools

Ghostifier: scans your inbox headers to find who holds your data, then unsubscribes, opts out of sales, and sends tracked deletion requests automatically. LINK

Replay QA Security Scan: runs automated penetration tests on web apps to catch injection flaws, broken access control, IDOR, and cross-tenant data exposure in AI-generated code. LINK

Servor: operational console for managing 2 to 50 servers through a web terminal, file editor, cron, and monitoring, with browser-signed commands and no inbound SSH port. LINK

Upsolve Row-Level Security: enforces data access permissions server-side, letting AI agents safely serve multiple tenants without ever seeing or bypassing row-level rules. LINK

AuthMonster: a two-factor authentication app with biometric locking, easy migration from Google Authenticator and LastPass, and a donation-based model instead of paywalls. LINK

Termaxa: a safety gate for AI coding tools that previews deletions, backs up files first, and blocks or prompts based on your policy, open source and written in Rust. LINK

Trending papers & reports

Water-system attack simulator gauges how often a hacked water network actually loses physical service, and a model trained to spot those cascades flagged real-impact cases far better than a scenario-only baseline across 2,160 synthetic trials. LINK

Source address spoofing gets a lightweight fix that lets routers verify a packet's claimed origin with a single quick check, blocking forged traffic without the setup costs that kept past defenses from being adopted. LINK

Over-powered tool picks by AI agents drop by up to ~19 points when each tool's access level is simply labeled, and paired with a caution instruction, safe task completions rise from ~19% to ~51%. LINK

Access control policies can be auto-generated and fixed by pulling permission rules scattered across code and docs, turning the ~41 to 56% of real projects with broken policies into nearly 100% valid ones. LINK

Login brokers let the middlemen that bundle Google and Facebook sign-in into one button handle authentication without ever seeing who you are or which sites you use, closing a major privacy hole in federated login. LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!