☕️ Fake Zoom updates give hackers control

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

🎭 Fake Zoom updates give hackers control

🍏 New macOS malware hijacks Chrome

📡 TP-Link flaws let hackers hijack networks

🔓 3 software flaws under active attack

🧩 Fake extensions steal developer data

Plus: 💡 6 strategies & tactics, 🎁 7 other news you might like, 🧰 6 tools, and 📚 5 papers.

🎭 Fake Zoom updates give hackers control LINK

A campaign called SMOKE#SCREEN is tricking users with fake Zoom updates that secretly install ConnectWise ScreenConnect, a legitimate IT remote-access tool, giving attackers persistent full remote control that looks like authorized IT activity.

Victims who run any of the initial files, spread through VBScript droppers, batch loaders, .NET executables and a polished fake Zoom HTML page that auto-downloads after two seconds, get a working ScreenConnect agent beaconing to attacker relay servers.

Securonix found the actor pivoted from destroying Microsoft Defender to evading it, adding a 180-second delay to break Elastic EDR correlation, and advises alerting on processes stopping WinDefend, adding C:\ as a Defender exclusion, or ScreenConnect connecting to raw IP addresses.

🍏 New macOS malware hijacks Chrome LINK

A new version of the XCSSET macOS malware, tracked as XCSSET v40, hijacks Google Chrome to spy on browsing, steal logins, and turn the browser into a hidden command channel, hitting developers who build infected Xcode projects.

Spreading through infected Xcode projects since April 2026 and affecting dozens of legitimate apps with thousands of users, the malware swaps Chrome's launch process for a wrapper that opens Chrome with remote debugging on and runs a component called chrome_remote.

Abusing the Chrome DevTools Protocol, chrome_remote injects JavaScript to capture credentials and API tokens, tamper with MetaMask transactions, and run shell commands via console messages as a fileless reverse shell; Google is working to extend Windows protections to macOS.

📡 TP-Link flaws let hackers hijack networks LINK

Forescout disclosed 15 vulnerabilities in the zero-touch provisioning (ZTP) systems of TP-Link's Omada networking ecosystem, warning that chaining some of the flaws could let attackers hijack entire fleets of managed routers, switches, and access points.

Chaining the new flaws with two earlier remote code execution bugs (CVE-2025-7850 and CVE-2025-7851), researchers showed an external attacker with no network access can abuse a race condition during cloud-based device adoption to intercept credentials and take over a user's cloud controller account.

Because one compromised controller manages a whole fleet, the attack chain could give root-level command execution on Omada devices; TP-Link has patched some issues but says fixes for structural weaknesses may not finish until later in 2026, and some low-severity flaws won't be patched.

🔓 3 software flaws under active attack LINK

CISA warned that hackers are actively exploiting three flaws in IBM Langflow OSS, N-able N-central, and Apache Tomcat, urging federal agencies to patch all three by August 7.

The most serious, CVE-2026-9198 in Langflow OSS (critical, CVSS 9.8), lets unauthenticated attackers run code by first grabbing a superuser token from an auto-login endpoint, then submitting malicious Python to a code-validation endpoint; a proof-of-concept appeared roughly a week after disclosure.

IBM patched the flaw on July 17 in Langflow OSS version 1.10.1, warning all default deployments are affected; CISA added it to its Known Exploited Vulnerabilities list on August 4.

🧩 Fake extensions steal developer data LINK

Security firm Manifold uncovered 77 counterfeit "evil twin" extensions on the Open VSX marketplace that impersonated legitimate developer tools while quietly collecting and transmitting data about the systems and development environments where they were installed.

The fakes reused the names, namespaces, and descriptions of real extensions but shipped from unrelated accounts with version 0.0.1; while 58 mainly sent the machine hostname, 19 gathered developer, Git repository, and continuous integration metadata about four to five seconds after activation.

The 19 reconnaissance extensions read the workspace's .git directory and enumerated up to 60 installed extensions plus GitHub, GitLab, and Azure DevOps identifiers, all reporting to mangorbit[.]com; the packages were removed by August 3, 2026, but Manifold says defenders must manually uninstall them and block that domain.

💡 Strategies & Tactics

> A few notes on AWS Nitro Enclaves: KMS integration: Explains how connecting AWS Nitro Enclaves to its Key Management Service creates new attack risks, and catalogs safeguards like hardcoding key IDs and validating encryption context.

> Securing Agentic AI Workflows in n8n: From Leaked API Keys to Encryption Key Compromise: A leaked n8n API key can chain into full encryption-key compromise, exposing every stored credential, so isolate instances by team, patch critical bugs fast, and keep admin access private.

> OWASP Subtractive Security Top 10 Project Released to Identify and Reduce Cyber Risks: OWASP's new framework prioritizes removing unnecessary access, services, and privileges to eliminate attacker paths entirely, treating detection as only a last resort.

> Public PoC Released for CUPS Vulnerability Allows Attackers to Gain Root Privileges: Public exploit code for a macOS printing flaw lets local users write files as root, so administrators should patch immediately before attackers weaponize it.

> 7-Zip Mark-of-the-Web Bypass Lets Malicious Files Evade Windows SmartScreen: 7-Zip's default setting strips the "downloaded from internet" tag off extracted files, so Windows SmartScreen skips its safety warning on malicious programs.

> 1-Click RCE Flaw in Cursor, VS Code, and Google Antigravity Exposes 50M Developers to Cyberattacks: A single click on a malicious link in AI code editors let attackers silently seize developers' machines because these tools share VS Code's vulnerable codebase.

Other news you might like

• Anthropic's Mythos created fake identities to fool humans in new cyber incidentLINK

• Massive supply-chain attack compromises 440 packages under four hoursLINK

• Prolific ransomware group behind SonicWall zero-day attacksLINK

• Bitcoin bridge Boltz suspends services as AI hacks outpace patchesLINK

• DarkSword Server Combines iPhone Exploits With Fake Apple ID Login PageLINK

• How legitimate cloud platforms enable phishers to bypass MFALINK

🧰 Trending tools

Perfai Security: automatically scans and fixes access control vulnerabilities in apps built with Replit, Lovable, Claude Code, and Cursor, making them production-ready in minutes.LINK

Constellation Gate AI: routes AI agent traffic through a gateway that blocks prompt injection, scans for secrets, logs audit trails, and cuts token costs 20-40% via compression and caching.LINK

Lunen.ai: an AI automation tool that logs every action taken, flags risky steps for approval, and gives security teams a full audit trail.LINK

HOL Guard: a firewall for AI agents that intercepts and blocks high-risk actions like deleting production data or exposing secrets before they execute.LINK

qsa.sh: runs a one-command external security scan of your server's public IP, using naabu, nmap, and nuclei to detect open ports, service versions, and known CVEs in about 30 seconds, no signup or storage required.LINK

Cynative Security Research Agent: an open-source AI CLI that answers plain-language security questions across cloud, code, and runtime, using read-only sandboxed scripts to prevent accidental infrastructure changes.LINK

📚 Trending papers & reports

Zero-knowledge fairness proofs let banks cryptographically prove a lending model's bias score and calibration to regulators on 32,768 real mortgage records, within 0.0029 accuracy, without revealing weights or customer data.LINK

Malware detection safeguards that flag when threats evolve can be undermined by the same tampering tricks used to fool the classifiers, but attackers must adjust their approach since these safeguards react differently than the classifiers they protect.LINK

AI recommendation systems built from chatting user and item agents become more vulnerable to manipulation as their interaction network grows denser, showing platforms must balance connectivity with security when automating recommendations.LINK

Speed shortcuts in AI chatbots, which skip unused parts of the model to run faster, leave a pattern that attackers can secretly watch and reverse to reconstruct users' prompts and answers with over 0.95 accuracy, even inside supposedly secure cloud hardware.LINK

Deepfake video detection gets a tougher test set of ~97,500 AI-generated clips, revealing that simple photo-based fake detectors, once their frame-by-frame guesses are combined smartly, beat top video-specific detectors, hitting 93.80% accuracy versus 79.99%.LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!