☕️ Microsoft patches nearly 400 security holes

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

🛡️ Microsoft patches nearly 400 security holes

🎥 AI finds Zoom flaw that hijacks devices

🎭 Russian hackers use fake job interviews

🔥 Firewall flaw lets hackers crash Cisco devices

🔍 Anthropic watermarks Claude text

Plus: 💡 4 strategies & tactics, 🎁 7 other news you might like, 🧰 6 tools, and 📚 5 papers.

🛡️ Microsoft patches nearly 400 security holes LINK

Microsoft's August Patch Tuesday fixed at least 398 security vulnerabilities across Windows and supported software, including one flaw already being actively exploited in the wild and two others that were publicly detailed before the release.

The actively exploited zero-day, CVE-2026-68820, is a privilege escalation weakness in the afd.sys driver that handles Windows socket connections on nearly every endpoint, letting an attacker who already has a low-privilege foothold from phishing take over the machine.

Rated 7.0 for high attack complexity because it relies on a race condition that must be triggered repeatedly until the timing lands, the flaw is not a break-in bug but a second step in a chain, and attackers are landing it anyway.

🎥 AI finds Zoom flaw that hijacks devices LINK

Researchers used publicly available AI models to build a Zoom exploit chain called ZOOMSDAY that lets an attacker take over another meeting participant's device just by being in the same call, with no clicks or downloads needed.

The zero-click chain, led by CVE-2026-53413 affecting Zoom Client v7.0.5 on Windows, macOS, iOS, and Android, abuses the annotation feature by sending specially crafted messages that corrupt the receiving client's memory to run attacker code and access the microphone or camera.

Zoom has released fixes and deployed a server-side filter for malicious messages, but that protection cannot inspect end-to-end encrypted meetings, so crafted messages can still reach vulnerable clients; organizations should update Zoom immediately.

🎭 Russian hackers use fake job interviews LINK

Russian state-linked hackers Sandworm are running fake job interviews to plant malware on IT staff, luring candidates from job platforms to Telegram and Zoom while impersonating firms like ATLAS Business Group and Sopra Steria.

Victims receive a trojanized WireGuard VPN client that hides a covert code-execution path: it accepts a non-standard SymmetricKey option carrying an AES-256-GCM payload and decrypts embedded PowerShell run through WireGuard's own script mechanism.

On Windows the PowerShell creates a scheduled task and pulls a second-stage payload from a URL, while the Linux version uses cURL to download an executable; CERT-UA says to hunt for unexpected scheduled tasks and unsigned WireGuard-derived binaries.

🔥 Firewall flaw lets hackers crash Cisco devices LINK

Cisco has patched an actively exploited zero-day in its Secure Firewall ASA and FTD software that lets attackers crash the appliances and knock them offline in a denial-of-service condition.

The flaw, tracked as CVE-2026-20349, lies in HTTP request processing and lets a remote, unauthenticated attacker force an appliance to reload by sending a specially crafted HTTP request to the Remote Access SSL VPN service.

Cisco became aware of active exploitation in August 2026 and urged customers to apply the available hotfixes; CISA added the bug to its catalog, instructing federal agencies to patch it by August 14.

🔍 Anthropic watermarks Claude text LINK

Anthropic is embedding hidden, machine-readable watermarks into text produced by newer Claude models, creating a provenance trail that can indicate content passed through Claude without being visible to human readers.

The watermark is applied at the model level so it travels when text is copied and may survive some editing, working across AWS, Google Cloud and Microsoft Foundry, and applies only to Claude models launched on or after August 2, 2026.

For files in SVG, PNG and JPG formats Claude attaches digitally signed provenance metadata using the C2PA standard, but Anthropic warns neither system is an AI detector since signals can vanish after paraphrasing, translation or extensive editing.

💡 Strategies & Tactics

> ShieldBreak: New Windows Zero-Day Bypasses Microsoft’s RoguePlanet Patch: A researcher publicly released working exploit code showing Microsoft's July fix for a Defender flaw fails, still letting attackers gain full system control on updated Windows machines.

> How Trail of Bits helps verify the integrity of your Signal chats: Trail of Bits runs one of three independent auditors that continuously verify Signal's public keys are consistent, so a compromised server can't secretly swap in an attacker's key.

> Chrome adopts what may be the best protection yet against account takeovers: Chrome now locks login session cookies to a device's security chip, blocking attackers who steal those cookies from reusing them to hijack accounts.

> Plug and Pwn Attack Abuses Windows PnP Drivers to Gain SYSTEM With Zero Clicks: Attackers can gain full Windows control by faking a USB device, since PnP (Plug and Play) driver installs run vendor code with SYSTEM privileges before login.

Other news you might like

• Mozilla updates GPG signing key for Firefox releases after exposureLINK

• Kimwolf botnet rebuilt to survive takedowns, researchers sayLINK

• Delta investigating after someone set up fake Wi-Fi network mid-flightLINK

• CopyEscape: Taking Over Docker Hosts with docker cpLINK

• Malicious SIMs can shut down phones, steal files, and drag 5G back to 2GLINK

• One ClickFix Lure Can Give Hackers a Persistent Remote Shell Through New CNCMachineRMS RATLINK

• DeadLock ransomware uses blockchain to resist infrastructure takedownLINK

🧰 Trending tools

FireTail: an AI security and governance platform that gives visibility into AI usage across your environments, helping teams detect and secure risks before they cause harm.LINK

BestDefense.io: continuously pentests every deploy, validates which vulnerabilities are truly exploitable, and auto-generates fixes so SaaS teams patch real risks fast.LINK

Sequirly: browser extension that scans prompts and file uploads before they reach ChatGPT, Claude, or Gemini, flagging API keys and personal data.LINK

DeepFrame: an authorized penetration testing studio for fast-moving web apps, delivering deep security assessments with clear reporting and follow-up retests.LINK

Origin: a confidential agentic stack combining a private LLM gateway, AI IDE, agents, attestation, and sandboxes for defense, finance, and regulated teams.LINK

Refuse: blocks known-vulnerable package installs across npm, pip, cargo, gem, go and 13 more managers before they hit disk, self-hostable via Docker.LINK

📚 Trending papers & reports

IoT security explanations can cost 700 seconds per batch for one detection model versus under 2 seconds for another, and smart triage still cuts compute by 15-32% without hiding dangerous false negatives.LINK

Tamper-proof execution tracking lets tiny embedded devices prove exactly what code path they ran, using hardware safeguards instead of vulnerable stored keys, while data sent back grows only in step with program size, not exponentially.LINK

Invisible code fingerprints let companies mark AI-generated code as their own without breaking how it runs, and let owners prove authorship later without ever exposing the secret watermark itself.LINK

Automated red teaming trains one chatbot to invent attacks against another, generating more effective English jailbreak prompts than existing benchmarks and, for the first time, doing the same in Turkish.LINK

AI agents that take real-world actions face far less security research than chatbots, a review of 85 studies finds attack research outpaces defense work nearly 4 to 1, with only 4.7% examining the risky tool-use and code-execution steps that actually cause damage.LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!