221M
Devices Vulnerable
6
CVEs Chained
3
Zero-Days Used
5+
Threat Actors
The Story
In early March 2026, Google's Threat Intelligence Group (GTIG), iVerify, and Lookout jointly published research on two iOS exploit kits: Coruna (linked to L3Harris/Trenchant, a U.S. government contractor) and DarkSword (a separate tool sold to multiple state-linked operators). Both use WebKit-based exploit chains to fully compromise iPhones and iPads through malicious websites. DarkSword was observed in active use by at least four distinct threat actors across campaigns targeting users in Ukraine, Saudi Arabia, Turkey, and Malaysia since November 2025.
On March 23, someone leaked a newer version of DarkSword on GitHub. The uploaded code is straightforward HTML and JavaScript with developer comments explaining each exploit stage. iVerify co-founder Matthias Frielingsdorf told TechCrunch: "This is bad. They are way too easy to repurpose. I don't think that can be contained anymore. So we need to expect criminals and others to start deploying this." He added: "The exploits will work out of the box. There is no iOS expertise required." A Google spokesperson confirmed they agree with this assessment.
A security hobbyist known as matteyeux demonstrated the point immediately, publicly hacking an iPad mini running iOS 18 using the leaked DarkSword sample and posting the results on X. The time from download to working exploit: minutes.
The threat escalation was immediate. By March 26, Russian threat actor TA446 (a Kremlin-linked espionage group previously tracked by Proofpoint) adopted the leaked DarkSword kit and began deploying it in spear-phishing campaigns. The emails spoofed the Atlantic Council and targeted Russian opposition figures, including Leonid Volkov, a prominent Russian opposition politician. Proofpoint noted that TA446 had never previously targeted iOS devices or iCloud accounts, but "the adoption of the leaked DarkSword iOS exploit kit has now enabled the actor to target iOS devices." This is the proliferation threat playing out in real time: a nation-state exploit becomes commodity malware, and within days, additional adversaries add it to their toolkits.
Apple responded by issuing emergency patches for older iOS versions and taking the unusual step of sending Lock Screen notifications to iPhones and iPads running vulnerable software, urging users to update immediately. Apple spokesperson Sarah O'Rourke confirmed the company was aware of the exploit and stated that Lockdown Mode also blocks DarkSword attacks. Despite these patches, approximately 14.2% of active iOS devices, roughly 221 million iPhones and iPads, remain on vulnerable versions according to iVerify's analysis of market data.
The Adversary Perspective
From a red team lens, DarkSword represents the most significant mobile exploit proliferation event of 2026. Three characteristics make it dangerous at scale.
The attack is entirely browser-based and requires zero interaction. The victim visits a compromised or attacker-controlled website in Safari. The exploit chain fires automatically. No clicks, no prompts, no app installs. The exploit begins with a WebKit/JavaScriptCore vulnerability (CVE-2025-31277 for iOS pre-18.6, or CVE-2025-43529 for iOS 18.6-18.7), achieves code execution, bypasses PAC via a dyld bug (CVE-2026-20700), escapes the WebContent sandbox into the GPU process (CVE-2025-14174), pivots from GPU to mediaplaybackd (CVE-2025-43510), and escalates to kernel (CVE-2025-43520) for full device control. The entire chain runs in seconds.
DarkSword is a smash-and-grab operation, not a persistent implant. Once it achieves kernel access, it rapidly harvests contacts, messages, call history, Safari browsing history, Wi-Fi passwords from the iOS keychain, SIM information, health records, notes, calendar databases, and cryptocurrency wallets. It exfiltrates everything via HTTP to an attacker-controlled server and disengages. There is no persistent implant to detect on the device afterward. The forensic footprint is limited to Safari browser history entries pointing to the exploit delivery domains.
The leaked code is plug-and-play. The GitHub upload includes developer comments, clear variable names, and documentation of each exploit stage. Frielingsdorf at iVerify described it as something anyone could host on a web server "in a couple minutes to hours." Lookout's Justin Albrecht said the leaked version "allows even unskilled threat actors to deploy the advanced iOS espionage kit, turning it into commodity malware." The barrier to entry has collapsed from nation-state capability to script-kiddie accessibility overnight.
Figure 01
- STEP 01
Victim Visits Malicious Page
Attacker hosts DarkSword on a web server or compromises a legitimate website (watering hole). Target visits the page in Safari on an unpatched iPhone/iPad. No interaction required. UNC6353 compromised a Ukrainian .gov.ua site for delivery. - STEP 02 // CRITICAL
WebKit RCE (CVE-2025-31277 or CVE-2025-43529)
JavaScriptCore JIT vulnerability exploited for remote code execution in the Safari renderer process. Different CVE used depending on iOS version. Achieves arbitrary code execution inside the WebContent sandbox. - STEP 03 // CRITICAL
PAC Bypass + Sandbox Escape
CVE-2026-20700 bypasses Pointer Authentication Codes in dyld. CVE-2025-14174 escapes WebContent sandbox into GPU process. CVE-2025-43510 pivots from GPU to mediaplaybackd. Three exploits chain to reach a privileged process outside the browser sandbox. - STEP 04 // CRITICAL
Kernel Privilege Escalation (CVE-2025-43520)
Kernel memory corruption vulnerability grants arbitrary read/write and arbitrary function call inside mediaplaybackd. Full device control achieved. Attacker can read and write any file, access any credential, and execute any code. - STEP 05
Data Harvesting (GHOSTBLADE)
GHOSTBLADE malware harvests contacts, messages, call history, Safari history, Wi-Fi passwords (keychain), SIM data, health records, notes, calendar, crypto wallets. Everything exfiltrated via HTTP to attacker C2. Entire process completes in seconds. - STEP 06
Clean Disengagement
No persistent implant installed. Exploit disengages after exfiltration. Minimal forensic artifacts remain on device. Only browser history entries pointing to delivery domains serve as detection indicators. Victim is unaware.
Exploit Chain CVEs & Campaign Details
- CVE-2025-31277 - JavaScriptCore JIT type confusion (RCE, iOS <18.6)
- CVE-2025-43529 - JavaScriptCore DFG GC bug (RCE, iOS 18.6-18.7) [ZERO-DAY]
- CVE-2026-20700 - dyld PAC bypass (user-mode auth bypass) [ZERO-DAY]
- CVE-2025-14174 - ANGLE memory corruption (sandbox escape to GPU) [ZERO-DAY]
- CVE-2025-43510 - iOS kernel memory management (GPU to mediaplaybackd)
- CVE-2025-43520 - iOS kernel memory corruption (privilege escalation)
- Malware: GHOSTBLADE (data harvester), GHOSTSABER (PARS Defense backdoor)
- Patched In: iOS 26.3, 18.7.3, 16.7.15, 15.8.7
- Vulnerable Population: ~221M devices (14.2% of active iOS devices)
- Known Operators: UNC6353 (Russia), UNC6748 (Saudi Arabia), PARS Defense (Turkey), TA446 (Russia, adopted post-leak)
- Leak Date: March 23, 2026 (GitHub)
- TA446 Campaign: March 26, 2026 (Atlantic Council spoofed emails, targeting Russian opposition)
Defensive Takeaways
- 1
Update Every iOS and iPadOS Device in Your Organization Today
This is not optional and it is not a routine patch cycle. A working, documented, government-grade exploit kit is publicly available on GitHub and being actively used by multiple threat actors. Update to iOS 26.3 or later, iOS 18.7.3 or later, iOS 16.7.15, or iOS 15.8.7 depending on the device. If your MDM platform supports forced updates, push them now. If it does not, send an urgent communication to all employees instructing them to update manually. Every unpatched iPhone in your organization is a one-click-away compromise waiting to happen.
- 2
Enable Lockdown Mode on High-Value Targets
Apple has confirmed that Lockdown Mode blocks DarkSword attacks. For executives, IT administrators, security personnel, finance team members, and anyone with access to sensitive systems or data, enable Lockdown Mode immediately. It restricts certain iOS features (message attachment types, web technologies, incoming FaceTime calls from unknown contacts), but for high-risk individuals, the tradeoff is worth it. This is the single most effective mitigation for users who cannot immediately update to a patched iOS version.
- 3
Treat This as an Enterprise Threat, Not Just a Mobile Threat
A compromised iPhone is not an isolated incident. DarkSword harvests Wi-Fi passwords from the iOS keychain, which can include corporate network credentials. It steals messages that may contain sensitive business communications. It grabs contacts that map your organization's relationships. If the device has email accounts configured, the attacker gets the credentials. If the device has VPN profiles, the attacker gets the configuration. A single compromised employee iPhone can become the entry point for a broader enterprise compromise. Treat unpatched mobile devices with the same urgency you would an unpatched VPN appliance.
- 4
Monitor for Watering Hole Indicators on Corporate and Partner Websites
DarkSword has been delivered via compromised legitimate websites, including a Ukrainian government domain. If your organization operates customer-facing websites, monitor for unauthorized script injection. Look for unfamiliar script tags loading external JavaScript, particularly files named
rce_module.js,rce_loader.js, or similar. If your employees visit partner or industry websites that could be targeted for watering hole attacks, consider blocking JavaScript execution from untrusted domains in your corporate browser policies. - 5
Check Devices for Compromise Using iVerify or Forensic Tools
iVerify is offering its Basic app for free until May to allow anyone to check their phone for DarkSword infection. For recent infections, the app's threat hunting feature can detect active compromise. Since DarkSword does not clean up Safari's browser history or WebKit-related databases, you can also use Mobile Verification Toolkit (MVT) or other forensic tools to search for known delivery domains in browser history. The file-based indicators are not backed up, so checking requires on-device access or a full filesystem dump. If any device shows indicators of compromise, rotate all credentials that were accessible from that device.