Europol kills three major malware networks

Hi there, this is your daily ☕️ Cyberpresso.

Europol dismantles three major malware networks LINK

• Europol and partners from nine countries took down the infrastructure behind three malware networks, Gootloader, CastleRAT, and StealC, in the latest Operation Endgame, hitting services used for ransomware, remote access, and credential theft.
• Authorities seized hundreds of servers and domains and recovered 6.4 million victim datasets, which will feed notification services like HaveIBeenPwned; that figure counts credentials or records, not necessarily distinct people.
• Gootloader spread fake legal documents from hacked WordPress sites to open the door for ransomware, so administrators are urged to change passwords, turn on multifactor login, and patch vulnerabilities to shut out the attackers.

Hackers hijack Asian telcos to spy LINK

• A spying group called SilentAir broke into mobile operators across at least six Asian countries to intercept the calls and texts of chosen people, mostly in Taiwan, says Stockholm telecom security firm Enea.
• SilentAir sent fake UpdateLocation messages over the trusted IPX network so that targets' calls and texts routed to its hacked equipment, and for some victims it grabbed two-factor codes right after diversion.
• Enea says the state-level group is disrupted for now but could return; it linked the activity to China only with low confidence and urged operators to treat the whole IPX as untrusted and add signaling firewalls.

Chinese hackers breach firms using AI LINK

• Chinese-speaking hackers broke into 10 organizations using AI agents run by the DeepSeek and Zhipu GLM models, which planned and carried out the attacks with barely any human help, according to analysts at Unit42.
• The attackers plugged two Chinese AI models into Cairn, a public open-source hacking tool, and just typed in a target domain and a one-line goal in Chinese; the model then built and ran a multi-stage plan by itself.
• Running from August to September 2026, the campaign hit semiconductor, defense and civil society targets through old, known flaws like weak passwords and public CVEs, leaking over a million member records; strong MFA, fast patching and watching single-IP traffic still help defenders.

Hackers Exploit Check Point Firewall Flaws LINK

• Hackers chained Check Point firewall flaws to gain root access on Israeli organizations starting October 30, then spread worldwide, planting backdoors and running a wiper that destroyed files on VMware ESXi hosts.
• Attackers broke in through vulnerable Mobile Access portals, then exploited CVE-2026-123039, a privilege flaw rated 7.8 in the Gaia operating system that lets any local user send root commands, hitting a couple dozen confirmed customers.
• Check Point released fixed Jumbo Hotfixes for affected releases R81 through R82.20; admins can stop the dynamic_server process as a stopgap and check that patched sockets show restricted permissions owned by root rather than open to all users.

New N-central exploit beats latest patch LINK

• N-able has locked down parts of its N-central platform after researcher Sean Bales revealed two new exploit sets, 4EverAble2Shell and Agent2Shell, that work against the newest version and let attackers run code as root.
• 4EverAble2Shell is already being exploited in the wild, chaining an Apache Axis flaw (CVE-2026-90436) with a privilege escalation bug to seize root, and one researcher counted 21 compromised servers across the US, UK, Germany, and other countries.
• Attackers are using the access to drop an XMRig Monero miner tuned to run on half the CPU cores to stay hidden, and N-able warns that Hotfix 2 alone isn't enough since 19 of the known victims were already running it.

Strategies & Tactics

> When an AI Agent Goes Wrong, Can You Stop It?: Stopping a rogue AI agent means a tested set of controls that revoke credentials, disable tools and block network paths, not just turning off the model.

> How attackers turned a Teams meeting booking into a device code phishing trap: Attackers disguised a Microsoft sign-in code as a Teams meeting booking reference, tricking an employee into approving the attacker's session and handing over account access.

> Ransomware at the hypervisor layer: how attackers use ESXi and vCenter features for malicious purposes: Attackers encrypt entire VMware virtualization setups using legitimate ESXi and vCenter management tools rather than exploits, because the hypervisor layer resists normal security monitoring.

> Your pinned MCP server may not be as pinned as you think: Pinning an AI tool server's version won't lock its underlying dependencies, which can silently swap in fresh code that carries your credentials unless you freeze everything.

> The XG Files: Falcon C2 Flying Under The (Browser) Radar: Palo Alto Networks' Unit 42 exposed Falcon, a browser-based attack tool that hides malicious code in trusted sites to steal credentials from government targets.

> Coding Agent Security for AppSec Leaders: Start With What Agents Can Reach: Secure what AI coding agents can reach, not just the code they write, because a stray credential lets one bad decision delete production in seconds.

Other news you might like

• US authorities freeze billions in stolen Binance crypto after massive North Korean hack LINK

• Researchers Discover 6 Zero-Day Vulnerabilities in Danfoss Refrigeration Controllers Allowing Remote Root Access LINK

• Citrix Patches 10 Critical NetScaler Vulnerabilities LINK

• Hackers Use Windows Named Pipes and SMB to Spread Malware Across Isolated Government Networks LINK

• Beast Ransomware Uses Hidden ‘666’ Control Panel to Encrypt Windows Systems and Network Shares LINK

Trending tools

Axari: an AI security teammate inside Slack or Teams that automates recurring tasks and workflows, freeing teams to focus on actual security outcomes. LINK

Servor: browser-based console for managing 2 to 50 servers with a terminal, file editor, cron, and monitoring using signed commands, no inbound SSH ports exposed LINK

Plango: monitors AI coding agents on your Mac, guarding secrets and files, blocking suspicious network activity, and scanning projects for vulnerabilities locally. LINK

face2social: reverse image search that matches a photo across Instagram, Facebook, LinkedIn, TikTok, and X to verify identities and spot scammers. LINK

AllModels: a unified OpenAI-compatible API for accessing abliterated models, letting you compare providers, latency, pricing, and capabilities through one endpoint. LINK

mcpgawk: a local guard that intercepts MCP tool calls, pausing any modified after approval until you review and permit them. LINK

Trending papers & reports

Model forgetting requests can be hijacked to secretly erase sensitive information a company never asked to delete, because related facts share overlapping memory, and the authors offer a defense that protects at-risk knowledge while honoring the original request. LINK

Europe's data-sharing rulebook fails to truly anonymize Google Search data regulators want shared with rivals, as the authors re-identified real users with simple, automatable attacks that meet none of the EU's three privacy criteria. LINK

Hidden text fingerprints let companies trace even short AI-generated passages back to their source, recovering a stamped ID almost perfectly in 200-word outputs and holding up against light paraphrasing, though very short or heavily rewritten text still slips through. LINK

Small bug-hunting models can rank which parts of open-source software to inspect first, helping find 187 memory-safety flaws, 12 rated critical, across 4,000 projects on commodity hardware without any expensive frontier AI. LINK

Disguised AI add-ons can pass safety audits at full quality yet secretly flip a model's decisions once it is compressed for deployment, so vetting add-ons only in their actual compressed form catches these hidden attacks. LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!