☕️ Adobe extension flaw exposed WhatsApp chats

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

🔓 Adobe extension flaw exposed WhatsApp chats

📌 Flaw lets hackers seize admin accounts

🐧 9-year-old Linux flaw hands users root

🛡️ Hackers exploit Check Point firewall flaw

Plus: 💡 4 strategies & tactics, 🎁 9 other news you might like, 🧰 6 tools, and 📚 5 papers.

🔓 Adobe extension flaw exposed WhatsApp chats LINK

A flaw in Adobe's Acrobat Chrome extension, installed in roughly 329 million browsers, could have let attackers silently steal a victim's WhatsApp Web chats, contacts, and account details just by luring them to a malicious webpage.

The bug, tracked as CVE-2026-48294 and described by Adobe as a UXSS-class cross-origin data disclosure flaw, was reported by Guardio and patched by Adobe in June, and did not require any WhatsApp vulnerability, malware, stolen logins, or device access.

In the attack, dubbed HermeticReader, a hidden frame abuses missing security checks in the extension's internal messaging system to feed it unverified commands, writing to local storage to activate Hermes, a dormant Adobe engine that bridges to WhatsApp Web and scrapes private data in plain text.

📌 Flaw lets hackers seize admin accounts LINK

Two critical flaws in the open-source phone system FreePBX let attackers remotely run code and hijack administrator accounts, affecting FreePBX versions 16 and 17 with a critical CVSS v4 score of 9.3, exploitable without any login or user interaction.

The main bug (GHSA-37j8-fhxx-9vhp) in FreePBX 17's User Control Panel before 17.0.9 lets unauthenticated attackers connect to the UCP Node server on ports 8001/8003 and inject Asterisk Manager Interface actions to run commands as the asterisk user, enabling full system takeover.

A second flaw injects SQL through crafted Caller ID names in SIP "From" headers to alter administrator credentials; Sangoma has patched both in UCP 17.0.9 and missed call modules 16.0.11 and 17.0.6, and advises firewall rules restricting UCP and SIP traffic filtering.

🐧 9-year-old Linux flaw hands users root LINK

A nine-year-old flaw in the Linux kernel's XFS filesystem, dubbed RefluXFS and tracked as CVE-2026-64600, lets an ordinary local user silently overwrite protected system files and take over a machine as root, even with SELinux enforcing.

The bug is a race condition triggered when two simultaneous O_DIRECT writes hit the same reflinked file, causing a stale reference check that corrupts on-disk blocks directly, letting an unprivileged user overwrite any readable file and gain passwordless root within seconds.

Present in every kernel since version 4.11 and estimated to affect over 16.4 million systems, it survives reboots and leaves no kernel logs; vendor-fixed kernels are available for RHEL, Oracle Linux, AlmaLinux, Rocky, and Fedora, and patching plus a full reboot is the only fix since no workaround neutralizes it.

🛡️ Hackers exploit Check Point firewall flaw LINK

Attackers are actively exploiting a critical authentication bypass flaw (CVE-2026-16232) in Check Point Security Management and Multi-Domain Security Management, the servers that push policy to Check Point firewalls, with a "handful" of customers already affected and notified.

The bug lets an unauthenticated attacker obtain an application login token and log in via SmartConsole with full admin privileges to change security policy and configuration, but successful remote exploitation requires internet access to the Management Server IP and no Trusted Clients restrictions.

Check Point released jumbo hotfixes for supported versions R81.20, R82, and R82.10, and advises limiting Trusted Clients and Management access to trusted IP addresses; CISA added the flaw to its catalog, requiring US federal agencies to fix it by July 25.

💡 Strategies & Tactics

> Flaws in Passkey Implementation Show Old Attacks Still Work: Flawed passkey implementations, not the cryptography itself, let attackers steal and reuse login credentials, so organizations must patch systems and validate carefully.

> Anthropic Launches Claude Security Plugin to Scan Codebases for Vulnerabilities Before Commit: Anthropic's new tool scans code for security flaws before developers commit it, catching vulnerabilities early inside the terminal without switching to separate software.

> Large-Scale GitHub Actions Abuse Powers a Distributed cPanel and WHM Exploitation Campaign: Attackers are abusing GitHub Actions runners to exploit vulnerable cPanel servers and steal sensitive credentials.

> A Hidden Line of Website Text Can Turn AWS Kiro Into a Remote Code Execution Tool: Hidden instructions on a webpage can trick AWS Kiro's AI assistant into running attacker code, proving user-approval prompts alone cannot secure AI coding tools.

Other news you might like

• New TrickBot Malware Variant Uses DNS Tunneling for Command-and-ControlLINK

• New msaRAT malware uses Chrome, Edge browsers to route C2 trafficLINK

• Federal agencies broaden alert on Iran-linked OT attacksLINK

• New Windows Stealer Uses AI Profiling to Identify High-Value Corporate VictimsLINK

• Hackers Clone Microsoft Login Portals to Capture Credentials and Session Tokens in Real TimeLINK

• Malicious NuGet Typosquat Targets Digitain Betting Platform and Rigs Game ResultsLINK

• Royal Ransomware Uses Qbot and Cobalt Strike to Rapidly Compromise Windows DomainsLINK

• OpenAI models escaped containment and hacked a major AI application libraryLINK

🧰 Trending tools

Perfai Security: automatically scans and fixes access control vulnerabilities in apps built with AI coding tools like Replit, Lovable, and Cursor, making them production-ready without security expertise.LINK

Playground: a sandbox for testing prompt injection techniques against AI agents, helping developers identify and understand security vulnerabilities before deployment.LINK

Astra Autonomous Pentest: an automated pentesting tool that runs 15,000+ security and compliance checks, combining automation with manual expertise to identify vulnerabilities before attackers exploit them.LINK

OpenBox: a trust platform that adds runtime governance, cryptographic verification, and compliance to agentic AI workflows via one SDK.LINK

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

Sequirly: scans prompts and file uploads in your browser to catch API keys, credentials, and personal data before they reach AI tools like ChatGPT or Claude.LINK

📚 Trending papers & reports

Phone malware scanners can shrink 3.5 times in size and run on just 0.0189 mJ per scan while still catching over 99.2% of Android malware, easing battery drain from security software.LINK

AI agent guardrails that predict how a task-performing bot's actions will play out before it clicks "execute" catch 15.9 percentage points more unsafe moves while letting 5.1 percentage points more legitimate tasks through.LINK

Jailbreak attack scoring now checks whether a chatbot's unsafe response actually delivers complete harmful content, catching real safety bypasses with 97.27% accuracy versus far less reliable existing evaluation methods.LINK

Twin Agent security design splits an AI assistant into a "scout" that reads risky outside content and a "worker" that takes real actions, blocking prompt injection attacks while keeping task performance high, unlike existing defenses that sacrifice one for the other.LINK

AI agent security testing automates spying on an AI assistant's own habits and tools first, then uses that intel to craft sharper attacks, exposing weaknesses standard security checks miss in real coding agents.LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!