☕️ iOS 26.6 patches nearly 90 iPhone flaws

Hi there, this is your daily ☕️ Cyberpresso.

In today's Cyberpresso:

🍎 iOS 26.6 patches nearly 90 iPhone flaws

🛡️ Microsoft launches first cyber AI model

🌐 Network flaw lets hackers run code

🦷 Dental insurer breach exposes 23M people

🤖 Botnet hijacks 200k devices worldwide

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

🍎 iOS 26.6 patches nearly 90 iPhone flaws LINK

Apple released watchOS 26.6, the sixth update to watchOS 26 from September, arriving over two months after watchOS 26.5 and delivering unspecified bug fixes along with security updates listed in Apple's release notes for the watch software.

Apple's release notes for watchOS 26.6 list only unspecified security updates, without naming specific fixed flaws such as ones allowing kernel-level code execution or root access on the Apple Watch, so the exact patched vulnerabilities stay undisclosed here.

The update installs free on an iPhone running iOS 26.6 through the Apple Watch app under General > Software Update, and requires the watch to have at least 50 percent battery while placed on a charger.

🛡️ Microsoft launches first cyber AI model LINK

Microsoft has released MAI-Cyber-1-Flash, its first AI model built exclusively for cybersecurity, trained on decades of data the company gathered while responding to hacking incidents affecting its customers across enterprise software, cloud, and operating systems.

Microsoft did not share the model with independent testers before release, instead claiming that once integrated into its security tools it would top the standard Cyber Gym benchmark after Monday's launch, outperforming OpenAI and Anthropic at roughly half the cost.

The model will also power Project Perception, which turns AI models into teams of autonomous agents that detect and remediate network vulnerabilities, with some Microsoft agents capable of imitating hackers to identify and patch security weaknesses on their own.

🌐 Network flaw lets hackers run code LINK

Arista has patched a maximum-severity flaw in on-premises VeloCloud Orchestrator (VCO), the platform used to manage VeloCloud SD-WAN deployments, that lets remote hackers run operating-system commands without any login and take over the host.

The bug, tracked as CVE-2026-16812 and rated maximum severity (CVSS 10/10), is being actively exploited in the wild and lets unauthenticated attackers with only network access to the VCO web interface reach privileged functions meant only for internal use, needing no tenant or operator credentials.

It is fixed in VCO versions 5.2.3.14, 6.1.3.4, and 6.4.2.4 and later, and while patching, admins should restrict the web interface to administrative networks, block three attacker IPs Arista shared, and review logs for unusual encoded web requests.

🦷 Dental insurer breach exposes 23M people LINK

Dental benefits administrator DentaQuest is notifying more than 23 million people of a data breach after hackers broke into its network in May 2026 and accessed personal and dental health information.

The intruders accessed data between May 17 and May 20, 2026, exposing names, addresses, Social Security numbers, Medicaid and Medicare IDs, and dental or vision details like diagnoses, treatments, and billing information for affected individuals.

The ShinyHunters extortion group claimed the breach, saying it stole 234 GB and published it after ransom talks failed; DentaQuest is offering 24 months of free credit monitoring, fraud support, and identity theft recovery.

🤖 Botnet hijacks 200k devices worldwide LINK

A botnet named Dysphoria has infected roughly 200,000 routers, cameras, and IoT devices worldwide, using them for distributed denial of service (DDoS) attacks and as traffic relays, according to QiAnXin XLab researchers who first spotted it March 25.

Dysphoria spreads through weak Telnet and SSH credentials and known device flaws, including React2Shell (CVE-2025-55182) plus older bugs affecting Huawei and DrayTek gear, while hiding its command servers via Ethereum ENS and Solana SNS blockchain domains to resist takedown.

Infected devices send a fixed 78-byte login packet and receive DDoS commands specifying duration, type, and targets; a late-June variant instead abuses UPnP to create 155 port forwarding rules, turning devices into network proxies, with operators claiming 4 Tbps DDoS capacity.

💡 Strategies & Tactics

> Critical vBulletin Flaw Lets Unauthenticated Attackers Execute PHP Code Remotely: A critical vBulletin bug lets attackers run malicious code without logging in, so forum operators should immediately update to version 6.2.2.

> Claude Cowork can escape its sandbox, rummage through all of your files: Researchers found Claude Cowork's local mode could exploit a Linux flaw to reach a Mac's whole filesystem, exposing private keys and credentials.

> AI-Assisted Research Uncovers Linux Kernel Zero-Day Enabling Root Privilege Escalation: Researchers used AI tools to speed up discovery of a Linux kernel flaw letting local users gain root access, making prompt patching essential.

> PeekList: How Brave’s Playlist bypassed FaceID Protection for Private Tabs: Brave's Playlist feature let anyone open locked private tabs without a Face ID or passcode prompt, since that shortcut skipped the authentication check entirely until Brave patched it.

> How an Unpatched N-Day Let Any Anonymous Visitor Rewrite WooCommerce Prices in ND Booking: A booking plugin let any anonymous visitor rewrite WooCommerce product prices because its only defense was a nonce published openly in the page source.

Other news you might like

• Hackers Pose as IT Helpdesk on Microsoft Teams to Deploy GoGRPC BackdoorLINK

• Mirage Kitten targets Middle East and Africa region with new malwareLINK

• Microsoft Defender for Endpoint leaves some Linux boxes defenseless after updateLINK

• Multiple FFmpeg Flaws Allow Arbitrary Memory Corruption via Malicious VideosLINK

• Five Progress LoadMaster Flaws Let Attackers Execute Commands and Gain Root AccessLINK

• MedusaHVNC Malware Uses Hidden Windows Desktops to Evade DetectionLINK

• Bank of Baroda probes major leak of customer data and internal records after dark web exposureLINK

🧰 Trending tools

Perfai Security: autonomously scans AI-built apps for access control, business-logic, and prompt-injection flaws, then auto-generates pull requests to fix them.LINK

MonoCloud for Startups: provides one identity layer for authentication, authorization, and API access control, letting startups secure users and services with fine-grained permissions for free.LINK

FireTail: an AI security and governance platform that helps you see and secure AI usage across all your environments, reducing shadow AI risk.LINK

Lunen.ai: an AI automation tool that logs every action taken and requires approval on risky steps, giving enterprises audit-ready transparency without sacrificing usability.LINK

Claudoscope: a free macOS menu bar app that browses Claude Code session history, tracks token costs, scans for leaked credentials, and lints CLAUDE.md configs.LINK

Let's Seal: an open-standard, self-hosted document signing tool providing free certificate-based signatures, transparency logging, and blockchain timestamping without vendor lock-in.LINK

📚 Trending papers & reports

Hidden chip malware can now be flagged before it ever activates, by reading subtle power-usage patterns to spot dormant hardware Trojans, closing the security gap left by tools that only catch attacks after they strike.LINK

AI coding assistants writing code from vague or incomplete instructions, the kind developers actually give, produce insecure code over 56% of the time, but adding security-focused wording to prompts cuts that risk by up to 45%.LINK

Self-driving cars sharing sensor data can be tricked by hacked vehicles into missing objects, but a new trust-scoring defense checks data from three angles to block even coordinated fake-consensus attacks.LINK

Chip power leaks can be traced automatically before a processor is even manufactured, pinpointing exactly which hardware signals and software instructions leak encryption secrets through power usage, letting chipmakers fix security holes before costly fabrication.LINK

AI safety blind spots get automatically mapped by tracing which internal "story settings" make chatbots drop their guardrails, boosting jailbreak success rates by up to 18.2 percentage points and working across GPT-5, Claude, and Gemini too.LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!