An ad can now infect phones

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

📱 Spyware infects phones via online ads

🎣 Microsoft takes down huge phishing network

🔥 Palo Alto firewall zero-day under attack

📧 Cisco email gateway zero-day gives root

🔑 Claude Code bug could leak SSH keys

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

Spyware infects phones via online ads

Source: CyberScoop

• Citizen Lab today revealed AdFection, the first technical proof of zero-click mercenary spyware delivered through targeted online ads, meaning a device can be infected the moment an ad loads, without the victim ever clicking anything.
• Israeli firm Discover-Tech poses as an ad company but runs a platform that lets surveillance partners, including the known vendor Candiru, fingerprint and infect devices as the ad renders; between March and June its ads appeared across over 120,000 websites and 70,000 apps.
• At least 45 devices at Al Jazeera were hit in attacks tied to Candiru, which can remotely take over a phone, steal data and access its microphone and camera; Google has removed Discover-Tech from its vendor registries and cut its ad-platform access, as has IAB Europe.

Microsoft takes down huge phishing network

Source: CyberScoop

• Microsoft, working with Japanese authorities, Health-ISAC and several security firms, dismantled an ecosystem of Chinese-language phishing kits and services, tracked as Storm-2990, that sent more than 1 billion phishing emails and texts to over 75,000 organizations this year.
• The kits, sold under names like YAM Store, CoGUI and Iron Man, impersonated banks, retailers and government agencies to steal logins and financial data, and abused the device code sign-in process to grab authentication tokens and access victims' accounts, mainly in US and Japanese healthcare, finance, logistics and IT.
• Under a federal court order Microsoft seized and redirected the infrastructure domains, disabled more than 20,000 YAM Store accounts and blocked over 100,000 impersonation domains; Resecurity found thousands of phishing templates across many countries, with over three-quarters of all messages aimed at Microsoft customers.

Palo Alto firewall zero-day under attack

Source: Cyber Security News

• Attackers are actively exploiting a zero-day in Palo Alto Networks PAN-OS that lets them run code with root privileges on affected firewalls, with the company confirming "very limited exploitation" and urging customers to apply fixes immediately.
• The flaw (CVE-2026-0322, high severity at CVSS 8.6) is a code injection bug in the management interface: an attacker already holding a high-privilege login and network access to that interface can run commands as root, no user interaction needed.
• It is not an unauthenticated path, and no workaround exists, so Palo Alto says the supported fixed releases, PA-Series, VM-Series, and Panorama deployments, with Cloud NGFW and Prisma Access unaffected, are the recommended protection; three attacker-linked addresses were published for reviewing management traffic.

Cisco email gateway zero-day gives root

Source: BleepingComputer

• Cisco has patched an actively exploited zero-day in its Secure Email Gateway that lets remote attackers with no login run commands as root on the underlying operating system.
• The flaw, CVE-2026-76519, is a relative path traversal bug in the gateway's Advanced Malware Protection feature that mishandles email attachments, so an attacker just sends an email with a crafted attachment to an unpatched device to write files and run code; it affects instances where File Reputation is enabled with older Content Scanner Tools versions.
• Cisco says attackers may erase traces after gaining root, so it urges checking firewall and network logs and, for customers it contacted, renewing credentials and cryptographic materials and restricting access to the appliance.

Claude Code bug could leak SSH keys

Source: CSO Online

• A bug in Anthropic's Claude Code Desktop app, called SneakPeek, let attackers steal a developer's SSH private keys, AWS credentials and GitHub secrets simply by asking the tool to review a malicious HTML file.
• Rated high severity (CVSS 8.7), the flaw abused the app's automated preview pane: HTML files over 256KB were written to disk and loaded as "file://" URLs, bypassing the script-blocking protections applied to smaller files and giving the renderer local filesystem access.
• Anthropic fully resolved the issue after an initial patch was bypassed, and is preparing a GitHub Security Advisory and a CVE; Noma advises keeping AI coding tools updated and using scoped, short-lived tokens instead of long-lived credentials on developer machines.

Strategies & Tactics

> From One Account to Every Mailbox: Breaking Zimbra's Identity Boundaries (Part 1): Rapid7 found over 80 Zimbra email-server flaws that let one compromised account impersonate senders, read others' mailboxes, and guess passwords without triggering lockouts.

> The Quiet Pivot - Weaponizing Client Configuration Files for Lateral Movement | HTB Blog: Attackers who control one machine can edit its SSH config file to silently run commands on any server the victim later logs into, needing no stolen credentials.

> Beyond valid credentials: How exposed AWS keys are tested for Amazon Bedrock access: Attackers now specifically test stolen AWS keys for access to Amazon Bedrock AI models, since those credentials fetch higher resale value and run up victims' bills.

> Penetration testing every deployment with AWS Continuum: Run an AI-driven penetration test scoped only to each code change as an automatic gate that blocks promotion to production when it confirms an exploitable vulnerability.

> GitHub - TheSecuredAnalyst/security-suite: Open-source security suite for OSINT, web scanning, API testing, SIEM integration, and AI: Bundles open-source reconnaissance, web and API security testing with local AI analysis so anyone can run authorized scans without cloud services or coding.

> The Blind Spots of Agentic Pentesting and How to Close Them: Pair creative AI-driven pentesting with faster repeatable tests and control-based validation so every vulnerability gets checked, even on systems attacks can't safely reach.

Other news you might like

• PoC Released for Critical Fortinet FortiClient EMS Authentication Bypass Vulnerability Cyber Security News

• UAT-6275 Hackers Exploit WinRAR Flaw to Target CIS Government Organizations Cyber Security News

• Hackers Backdoor Nextcloud Servers Through Critical Flaw in Collabora’s Built-in CODE App The Hacker News

• CikaBika Bot RAT Uses Discord for CC SecurityWeek

• SonicWall NSM Bugs Let Attackers Intercept Communications and Execute Commands GBHackers

• Fake Chrome App on Google Play Evades Detection and Reaches 100,000+ Downloads Cyber Security News

Trending tools

Decawork: brings employee-built AI agents under IT control, moving them onto company accounts and managing their access, oversight, and retirement from one place. decawork.ai

VibeDefend: enforces security and business rules inside AI coding agents like Cursor and Copilot, catching injections, malware, and policy violations before the PR. cybedefend.com

Google Gemini 3.8 Flash and Cyber: fast, low-cost models for agentic coding, multi-step reasoning, autonomous tasks, and cybersecurity work like vulnerability detection. Google Blog

Replay QA Security Scan: pentests your web app on every QA pass, catching injection flaws, broken access control, and IDOR with triaged bug reports and suggested fixes. replay.io

chat-recall: searches your team's AI assistant conversations across six coding tools, strips passwords locally, and surfaces past work so you avoid redoing it. chatrecall.dev

Koreshield: inspects customer input, retrieved context, and proposed actions in AI support workflows, blocking untrusted content from becoming trusted instructions with detect and enforce modes. koreshield.ai

Trending papers & reports

Malicious agent skills can be caught before they run by a lightweight scanner that flags nearly every attack while wrongly flagging only 20-25% of safe automations, far fewer false alarms than existing filters. arXiv

Drone autopilot auditing uses a team of AI agents to automatically hunt for subtle logic flaws in open-source flight software, catching unsafe behaviors that don't trigger crashes and that older testing tools miss. arXiv

Adversarial patch testing attacks each software security fix with new variations of the same vulnerability, catching a hidden exploit in 41% of cases and raising verified fix rates from ~78% to ~85%. arXiv

Live code security checks catch flaws while an AI drafts code in its reasoning, prompting fixes mid-stream and lifting secure, working output from under 15% to as high as ~98%, using fewer tokens than after-the-fact repair. arXiv

Node.js security testbed gives researchers a continuously updated collection of 1,159 working exploits for nearly 1,000 real JavaScript packages, replacing stale datasets that stopped in 2022 and revealing that flawed packages are getting much larger. arXiv

See you tomorrow for a new dose of ☕️ Cyberpresso!