Hi there, this is your daily βοΈ Cyberpresso.
In today's Cyberpresso:
π¨π³ Chinese hackers exploit VMware zero-day
π Attackers exploit Cisco Expressway flaw
π©Ή F5 patches critical BIG-IP flaws
π₯οΈ Linux VM flaw exposes host server memory
Plus: π‘ 6 strategies & tactics, π 7 more stories you might like, π§° 6 tools, and π 5 papers.
π¨π³ Chinese hackers exploit VMware zero-day LINK
- Broadcom has patched a critical authentication bypass in VMware vCenter Server that a suspected Chinese state-sponsored group, Armory Castle, has exploited in zero-day attacks since September 2026 to seize control of virtualized environments.
- The flaw, CVE-2026-123890, lets an unauthenticated attacker with network access gain administrator rights over vCenter and the virtual machines it manages; Broadcom released an emergency fix yesterday but offers no official workaround.
- Mandiant found the hackers used a custom tool called BONESAW to create an admin account and steal vCenter's directory database and SAML signing keys, then forged tokens to bypass multifactor login and reach ESXi hosts.
Attackers exploit Cisco Expressway flaw LINK
- Attackers have exploited a critical Cisco Expressway flaw, CVE-2026-76466, to breach a Russian organization shortly after a public proof-of-concept appeared, gaining root-level command execution on internet-facing devices and pivoting into the internal network.
- The bug, rated 9.8, lets an unauthenticated remote attacker run operating-system commands as root because of persistent cookies and weak client-certificate checks; Cisco released patches on October 21, 2026, and urges users to update immediately.
- Russian firm F6 said the attackers reached two cluster nodes, then ran custom Python scripts to scan the network, probe SMB, SQL Server and WinRM, and steal credentials from GitLab, with the scripts possibly written using AI tools.
F5 patches critical BIG-IP flaws LINK
- F5 has told BIG-IP customers to patch five critical vulnerabilities, led by CVE-2026-121805, that let unauthenticated attackers crash the system's traffic engine or run their own code remotely on affected setups.
- The stack-based buffer overflow flaws hit the BIG-IP Access Policy Manager and the iRules scripting feature; F5 said there is no workaround, so teams must install the out-of-band hotfixes it released on Wednesday.
- None of the five are known to be exploited yet, but F5 bugs are frequent targets, with Shadowserver tracking over 177,000 exposed BIG-IP instances and CISA flagging seven F5 flaws as actively attacked since November 2021.
Linux VM flaw exposes host server memory LINK
- A flaw in KVM, the hypervisor built into the Linux kernel, let researchers read about 127 GB of a cloud server's memory from inside their own virtual machine on DigitalOcean, where it can include other customers' data.
- Tracked as CVE-2026-99312, the bug hits only AMD hosts with nested virtualization on; there is no known attack or public exploit code, and fixed kernels shipped upstream on Tuesday, with DigitalOcean live-patching its fleet.
- A pointer left aimed at freed memory let a guest forge the table that maps its addresses to the host's RAM; operators running their own AMD KVM hosts must update the kernel and reboot, or turn nested virtualization off.
Strategies & Tactics
> Detecting Timeroasting the Hard Way: Read a protocol's own binary code rather than attacker tools to build network-traffic detection for timeroasting attacks that leave no login records.
> .:: Abusing the Windows Smart Card Minidriver Architecture ::.: Malware can silently intercept a Windows smart card's PIN and remotely perform signing operations by hooking the minidriver layer that handles all card types.
> Hardening GitHub Actions Against Insider Threats: Shrink the damage from stolen or insider GitHub credentials by requiring phishing-resistant security keys, killing long-lived tokens, and splitting release approval across separately protected accounts.
> What Shadow AI Looks Like on Your Network: Monitor network traffic for unusual outbound data spikes and new destinations to spot unauthorized AI tool use that approved-software lists miss.
> The New DDoS Economy: Bigger Attacks, Better Sources, Lower Barriers: Junk-traffic attacks that overwhelm websites now use real-looking home internet addresses and precise targeting, forcing defenders to filter selectively rather than just block bandwidth.
> Backdooring LLMs - Do you trust your AI provider?: A hidden trigger phrase can make a downloaded AI model silently run attacker commands, so sandbox agents and monitor tool calls since hosted models can't be inspected.
Other news you might like
- New Unpatched Windows 0-Day FastUnlock Lets Someone Holding a Locked PC Run Programs on It LINK
- CyberPower PowerPanel Exploited to Run Malicious Code on Domain Controller LINK
- Spying on Governments with Zero Lines of Malware: A Custom SoftEther Bridge in Webwormβs Toolkit LINK
- React2STEAL: Hijacked GitHub Accounts Exploit React2Shell and Publish Stolen Secrets LINK
- Models Can Communicate via Inference Servers Despite Being Sandboxed LINK
- crypt0p3g/pass-the-device: Pass-the-Device (CODE BLUE 2026): slides (EN/JA), recon tools and experimental Sigma/KQL detections for VPN/SASE device trust after endpoint compromise. LINK
- BADBOX 3.0 malware used Android TV boxes to generate 40 billion daily bid requests LINK
Trending tools
Fraud Blocker: detects and blocks bots, click farms, and fake clicks on your ad campaigns, so you only pay for genuine traffic and leads. LINK
VibeDefend: embeds security and business rules into AI coding agents like Cursor and Copilot, catching vulnerabilities and malicious packages before code reaches your PR. LINK
Servor: browser-based console for managing 2 to 50 servers with a terminal, file editor, cron, and monitoring, using signed commands so no inbound SSH ports stay exposed. LINK
face2social: reverse image face search that scans Instagram, Facebook, LinkedIn, TikTok, and X to verify identities, flag mismatched profiles, and catch scammers. LINK
Upsolve Row-Level Security: enforces data access permissions server-side, letting AI agents safely serve multiple tenants without bypassing or exposing row-level rules. LINK
mcpgawk: a local guard that intercepts MCP tool calls, pausing any that changed after approval until you review and permit them. LINK
Trending papers & reports
AI agent forensics is a new tool that lets security analysts trace a breach back to its exact entry point with just a few queries, bridging the gap between high-level activity logs and underlying code. LINK
Image-based decision bots can be reliably tricked by tampered pictures, with doctored images fooling fast structured-decision systems up to ~70% of the time, exposing a security gap for companies automating choices from visual inputs. LINK
Automated hacking for smart devices uses a team of AI agents to scan, break into, and take control of connected gadgets like smart cameras and plugs on their own, speeding up security testing of IoT networks. LINK
Image-generator safeguards can be tricked into producing banned content by splitting a harmful claim into a question plus a check mark, letting gpt-image-2 generate ~43% of such requests versus ~16% for direct statements. LINK
Drone swarm spoofing can be caught even by sophisticated GPS attacks that fake an entire formation's location, by spotting tiny timing differences across drones, detecting threats in every turn window versus at most 5% before. LINK
See you tomorrow for a new dose of βοΈ Cyberpresso!