Friday 06 November 2026 | Join Free | Upgrade

Together with

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

🏛️ Traffic routed through White House network

❄️ Hackers overheat supermarket freezers

🔓 Anthropic won't fix Claude code flaw

📌 OpenSSH flaw lets intruders bypass login

🐧 Backdoor runs Linux commands silently

Plus: 💡 5 strategies & tactics, 🎁 6 more stories you might like, 🧰 6 tools, and 📚 5 papers.

Faradex.ai is a fully private AI workspace powered by Claude, deployed in seconds.

Your prompts and data are never stored or used for training.

Two-week free trial and $150 in AI credit for qualifying firms.

Start your two-week free trial →
🏛️ Traffic routed through White House network LINK
  • Security researchers found that residential proxy services routed internet traffic through the White House network, known as AS2650 and registered to the Executive Office of the President, meaning anyone could send requests that appeared to come from it.
  • The team's probes exited through the White House network 236 times from a handful of addresses on most days of the month, with 28 separate proxy services able to use it, confirmed independently by detection firms Spur and Synthient.
  • Most activity happened on weekday evenings Eastern time, hinting at a person whose device runs an infected proxy SDK; such flawed SDKs have spread botnets like Kimwolf, which reached about 2 million devices.
❄️ Hackers overheat supermarket freezers LINK
  • The U.S. government is warning of an ongoing threat after hackers raised temperatures on internet-connected supermarket freezers, cold storage units, military fridges, and a blood bank, spoiling food and threatening critical infrastructure worldwide.
  • CISA, the FBI, and other agencies said two waves of attacks hit refrigeration units across at least a dozen countries, including remote-controlled freezers made by Danish firm Danfoss, which had stopped sending its customers security updates.
  • Officials urged affected industries to disconnect remote control systems from the internet, strengthen passwords, apply patches, and retire unsupported equipment; the U.S. hasn't named a culprit, but researchers broadly blame Iran.
🔓 Anthropic won't fix Claude code flaw LINK
  • Anthropic will not patch a design flaw in its Claude Dispatch feature, instead choosing to retire the tool by the end of October 2026, leaving the problem open to attack until that shutdown date.
  • Dubbed ThisBash by Forever Security, the attack lets a malicious browser extension send commands through a logged-in Claude session to Claude Desktop, even on a separate machine, with a timer auto-clicking every "Allow" prompt.
  • The proof of concept needed no browser exploit and quietly opened Calculator, but a real payload could steal files or plant backdoors; researchers say Claude Code Remote Control and Cowork sessions remain open to the same trick.
📌 OpenSSH flaw lets intruders bypass login LINK
  • Attackers are already abusing an OpenSSH bug (CVE-2026-123158), nicknamed "sshole," that lets remote intruders log into passwordless accounts on Alpine Linux and other musl-based systems without knowing any password at all.
  • Security firm Previdian saw the first attack less than nine hours after the flaw went public, confirmed root access on an Alpine honeypot, and counted eight attempts from one Japan-based address using oversized passwords.
  • OpenSSH 10.6p2, released yesterday, fixes it, and Alpine patched all branches the same day; admins should restart sshd or set PasswordAuthentication=no, since unverified proof-of-concept exploits are already on GitHub.
🐧 Backdoor runs Linux commands silently LINK
  • Researchers at QiAnXin's XLab found a backdoor called cwpdoor hiding inside the web server for Control Web Panel, which runs Linux commands sent through web requests that never show up in the server's access logs.
  • The backdoor is a rebuilt version of the cwpsrv program that acts normally until a request arrives with a valid signature in custom headers starting with X-Ngx-Integrity, and it then runs the attached command.
  • XLab says real victims already exist, after an infected copy was uploaded from Brazil on September 30, and it recommends restoring the original cwpsrv, upgrading, and changing all control panel passwords since it had zero detections.

ManageEngine gives IT teams complete control over endpoints, networks, applications, and security. From one integrated console.

• Unified endpoint management across Windows, macOS, Linux, and mobile

• Network monitoring, ITSM ticketing, and automated patch management

• AI-powered analytics and compliance reporting

Find your it solution

💡 Strategies & Tactics

> ChatGPT lockdown mode bypass: phishing, malware delivery, and silent data exfiltration: Check Point found a flaw letting attackers slip past ChatGPT's Lockdown Mode to steal chats and credentials, proving built-in AI guardrails aren't foolproof.
> Even gem list can run someone else's code: Installing a Ruby gem can silently run an attacker's code through build scripts or plugins, so disable those behaviors and check projects for the four data-stealing gems.
> Virtualization security: attack layers and misconfiguration impact: Ransomware crews now seize entire virtual server farms through reused root passwords and overprivileged monitoring accounts, so patch and harden configurations together.
> You Had Me at "Hello": Attackers exploit an old flaw in the open-source AI platform Flowise, disguising code-injection commands as "hello" messages to steal cloud credentials and pivot into company accounts.
> Barracuda networks is your new best friend (Barracuda Email Security Gateway pre-auth remote root RCE CVE-2026-116606 / CVE-2026-111560): Chaining two Barracuda email appliance flaws lets an unauthenticated attacker bypass login and inject commands as root, risking full takeover of the gateway guarding corporate mail.

Other news & articles you might like

  • mSecure password manager: Secure your credentials after an IT incident LINK
  • Microsoft paint 0-click vulnerability could allow hackers to take control of Windows PCs LINK
  • Vercel rolls out new protection against stolen session cookie attacks LINK
  • CPython flaw exposes AI inference servers to sandbox escape attacks LINK
  • Hacking on autopilot: Inside ARTEX cybercrime campaigns LINK
  • OpenAI model wrote its own network code to get around sandbox restrictions LINK

🛠️ Trending tools

Harden: a free, local tool that inspects AI coding agent commands before execution, blocking dangerous actions while keeping everything on your machine. LINK
Replay QA Security Scan: runs automated penetration tests on web apps, catching injection flaws, broken access control, IDOR, and cross-tenant data leaks in AI-generated code before production. LINK
Servor: web console for managing 2 to 50 servers via terminal, file editor, cron, and monitoring, using browser-signed commands with no inbound SSH port. LINK
Upsolve Row-Level Security: enforces data access permissions server-side, so AI agents can safely serve multiple tenants without seeing or bypassing row-level rules. LINK
Stroq: a local audit tool that tracks and explains the actions your coding agent took and the reasoning behind them. LINK
Koreshield: screens customer messages, retrieved documents, and tool calls before your AI agent acts, catching data leaks, hidden instructions, and unsafe actions while logging every decision. LINK

📚 Trending research papers

Agent memory tampering reveals that AI assistants which learn from past interactions can be corrupted through five entry points, so even genuine successful experiences can produce unreliable behavior once stored, exposing a security gap for memory-based products. LINK
Quantum-safe encryption rollout gets its first full review across the software building process, finding that current tools prove the new standards technically work but leave big gaps in actually deploying them safely at scale. LINK
Agent injection defense blocks hidden instructions from redirecting an AI assistant's emails, files, or payments to attackers by checking that action targets came from trusted typed tool fields, stopping every standard attack tested while keeping task performance highest. LINK
AI agent handoffs can lose or corrupt pieces of a user's request as subagents pass work along, but digitally sealing a checklist of requirements cut unsafe results under attack from ~69% to ~9%. LINK
AI agent hijacking can happen not just by hijacking an agent's whole task but by subtly tampering with how it decides within a legitimate task, a sneakier attack that current defenses miss, leaving up to ~12% success. LINK

🎓 Want to master the AI tools we cover every day?

Our AI Academy has 330+ step-by-step tutorials on ChatGPT, Claude, Perplexity, and every tool that matters. No fluff — just practical workflows you can use at work. Try it free for 7 days.

💬 How did you find today's edition?

We read every reply — just reply to this email and let us know how we can improve!

★★★★★  Nailed it
★★★  Average
★  Fail

Not subscribed to ☕️ Cyberpresso yet? Subscribe for free

Keep reading