Microsoft just patched a record 1,413 flaws

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

🪟 Microsoft patches record 1,413 flaws

🔓 Stolen login exposes some Meta source code

🇰🇵 DOJ charges 14 in North Korea IT scheme

🔑 Ivanti VPN zero-days grant root access

🤖 AI builds $250K Chrome exploit

Plus: 💡 8 strategies & tactics, 🎁 7 more stories you might like, 🧰 6 tools, and 📚 5 papers.

🪟 Microsoft patches record 1,413 flaws LINK

• Microsoft set a new Patch Tuesday record yesterday, shipping fixes for 1,413 vulnerabilities across Windows, Office, SharePoint and other products, one of which is already being exploited in attacks seen in the wild.
• The exploited bug (CVE-2026-96951, CVSS 8.8, high severity) is a command injection flaw in Windows Shell that lets attackers slip past Mark of the Web defenses by tricking a user into opening a specially crafted file sent via email or message.
• CISA added the flaw to its Known Exploited Vulnerabilities catalog, requiring federal agencies to patch by Friday; Microsoft also detailed a publicly known .NET flaw that lets a logged-in attacker run unauthorized code by bypassing Windows Defender Application Control.

🔓 Stolen login exposes some Meta source code LINK

• Hackers used an employee's stolen login credentials to break into some of Meta's GitHub repositories, taking product source code and personal information tied to roughly 1,500 people, mostly Meta employees in the United States.
• Meta said the attackers got in through stolen credentials rather than a flaw in a Meta product, and its internal systems were not breached; about 99% of the stolen material was software code, with personal information making up the rest.
• Meta detected the intrusion several weeks before confirming it on October 13, 2026, blocked the attackers, rotated secrets in the affected repositories, added scans to catch mistakenly uploaded user data, and is offering affected people credit and dark web monitoring.

🇰🇵 DOJ charges 14 in North Korea IT scheme LINK

• The Justice Department has charged 14 people and seized about 187 computers in Operation Lonestar Roundup, a crackdown on North Korea's scheme to place IT workers in American companies using stolen or fabricated identities.
• Two U.S. citizens and 10 others in China allegedly held remote jobs at more than 70 companies from 2019 through 2025, with facilitators running "laptop farms" that kept employer-issued laptops online with remote-access software for overseas workers.
• The FBI warned the scheme now pays U.S. citizens to sit interviews and host laptops and sometimes threatens former managers, and since May the group has spoofed device identifiers to make a video capture card look like a webcam, which CrowdStrike says makes a lasting takedown unlikely.

🔑 Ivanti VPN zero-days grant root access LINK

• Attackers used a chain of Ivanti Connect Secure VPN zero-days to gain unauthenticated root access, steal credentials, and evade detection, hitting a fully patched appliance running version 22.8R2.4 before anyone noticed.
• The flaw (CVE-2026-104935, high severity at CVSS 8.6) lets an attacker bypass the login page by passing a URL check while an HTTP header redirects to a protected internal script, then win a race to run an uploaded Perl script before it's deleted.
• Ivanti patched two of the three bugs on October 13, with the privilege-escalation fix still coming; Volexity warns the on-disk Integrity Checker Tool missed the attack because changes were made in memory and restored.

🤖 AI builds $250K Chrome exploit LINK

• Security firm XBOW used AI agents running OpenAI's GPT-5.6 Sol model to build a full-chain Chrome exploit in six days for about $40,000 in computing costs, earning a $250,000 reward from Google, which has fixed the flaws.
• The chain linked three bugs to break out of Chrome's renderer sandbox: one in the V8 JavaScript engine's Maglev compiler gave memory access inside V8, an out-of-bounds write in the animated WebP decoder left the V8 sandbox, and a Crashpad crash-reporting flaw on Windows caused memory corruption outside the sandbox.
• Disclosed on October 13, 2026, the demonstration targeted Windows with a crafted page launching a program on a test system and was not used against people; XBOW estimated the issue could affect about one billion users, and Google has fixed all three flaws.

💡 Strategies & Tactics

> Vghost: Beyond the Privilege Boundary · Tencent Zhuque Lab: A decade-old Linux flaw lets local users gain root or guests escape their virtual machine, so patch the kernel and restrict vhost device access immediately.

> Visual Studio Copilot and SSMS Remote Code Execution (CVE-2026-103572): A single trailing dot in a filename let Copilot in Visual Studio and SQL Server Management Studio skip security checks and run malicious commands, now patched by Microsoft.

> A UAF turned RCE: A memory bug in curl's HTTP/2 uploads let a malicious server run its own code, prompting a design shift from pointers to safer transfer lookups.

> When LLM Gateways Go Wrong: Findings from LiteLLM: A missing permission check on LiteLLM's config endpoint let any logged-in user seize admin control, proving AI gateways still need basic access controls.

> Four-Byte Write and Timing Bug Let Researchers Run Code in Windows SmartScreen Enclave: Researchers broke into a sealed Windows memory area meant to protect passwords, showing that signed but flawed enclaves can still be loaded to hide malicious code.

> (Post-Quantum) Cryptographic Right Answers: Use hybrid post-quantum algorithms for all public-key cryptography now, because quantum computers threaten to break today's encryption by 2029.

> A story of wifi secrets, UART lockdowns and a damaged flash memory chip: Researchers extracted a smart-home hub's firmware by desoldering its chip, then reflashed a modified recovery image to bypass the serial-port lockout and gain root access for vulnerability hunting.

> Hands Off the Keyboard: How AI Agents Run Real Intrusions / breachcache: Autonomous AI agents now run full hacking intrusions, picking targets, finding exploits, and stealing data, with no human at the keyboard, meaning attacks scale faster.

Other news you might like

• iRhythm Data Breach Impacts at Least 360,000 People LINK

• SAP warns of maximum severity OAuth2 SSRF security flaw LINK

• Docker Hub flaw let attackers impersonate users, tamper with images LINK

• Ghostscript flaw lets hackers hijack servers via PDF uploads LINK

• Hirudo Android Trojan Uses Patched Banking Apps to Bypass Their Built-In Security Checks LINK

• Hackers Abuse Google Search to Attacks ITadmins with Weaponized PuTTY Installer LINK

• Critical FortiSandbox GUI Vulnerability Allows Remote Attackers to Bypass Authentication LINK

🧰 Trending tools

Harden: a free, local security tool that vets AI coding agent tool calls before they run, keeping your repo and output on your machine. LINK

CtrlOps: audits server security across 25 checks, manages access and roles fleet-wide, searches logs, and deploys GitHub apps, all locally without touching the terminal. LINK

Axari: an AI security twin that connects your tools and teams, taking goals or recurring tasks through Slack or Teams until completed. LINK

Servor: manages all your servers from one console with a web terminal, file editor, scheduled jobs, monitoring, and alerts, no open ports required. LINK

mcpgawk: monitors your agents' MCP server calls locally, holding any tool that changed after approval until you review it, with no uploads. LINK

Strac Comply: automates SOC 2, ISO 27001, HIPAA and PCI DSS compliance by connecting your cloud tools, running 100+ continuous tests, and mapping AI apps to sensitive data. LINK

📚 Trending papers & reports

Payment guardrails for AI agents cap how much money autonomous agents can lose to fake service providers in open markets, so that scammers spinning up cheap fake identities and reviews cannot inflate their trustworthiness to drain honest clients. LINK

Image watermarking for AI art hides an invisible provenance tag in generated images that survives the strongest removal attacks, keeping over 88% of the tag readable where rival methods fail, aiding copyright and misuse tracing. LINK

Vulnerability patch triage ranks which security flaws actually matter by checking whether they are reachable in your specific setup, finding that most of 6,138 flagged flaws across 30 apps pose no real risk. LINK

AI agent attacks show that hidden malicious instructions can trick autonomous assistants into leaking sensitive data and spreading like infections between connected agents, exposing serious security gaps as businesses deploy them in real workflows. LINK

Power sensors on phone chips can be read by ordinary apps to steal memory data and secret bytes without any timing tricks, cracking devices in minutes even when traditional defenses hold. LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!