BREAKING NEWS: CyberCartel uses Grok, Kimi, & Claude models to attack LATAM banks

Please complete this form for your free AI risk assessment.

How the Optimizely Compromise Routed iPhones Toward the Coruna and DarkSword iOS Exploit Kits

Malicious Optimizely scripts tried to route iPhones to the Coruna and DarkSword exploit kits. See what researchers confirmed and what remains unproven.

Product & Company News

Loading audio player...

Ben Actis
November 17, 2026

# min read

Coruna and DarkSword iOS Exploits in the Optimizely Attack

On November 16, an attacker injected malicious JavaScript into the customer projects of Optimizely's Web Experimentation platform, and for iPhone and iPad visitors that code was built to load either Coruna or DarkSword, two iOS exploit kits that Google's Threat Intelligence Group documented earlier this year. Report URI found the code in 188 snippets belonging to 178 organizations out of 1,611 snippets it checked, including CNN, the BBC, Zoom, and Salesforce. Researchers have confirmed the script delivery and the routing of iOS visitors toward the kits, but no successful iPhone compromise has been publicly verified, and Report URI says it cannot see how many iOS devices were infected before the attack was stopped.

We think the delivery path is the most concerning part, because capabilities that began in targeted surveillance and espionage are now being routed through a trusted dependency on sites with very large audiences.

How Coruna and DarkSword iOS Exploit Kits Spread Beyond Their Original Operators

Google first captured pieces of Coruna in February 2025, when a customer of a commercial surveillance vendor was using the exploits inside a previously unseen JavaScript framework. By summer 2025 UNC6353, a suspected Russian espionage group, was loading the same framework through hidden iframes on compromised Ukrainian websites for selected iPhone users. At the end of the year Google found the complete kit on a very large set of fake Chinese finance sites run by UNC6691, a financially motivated actor, which delivered it to iOS visitors regardless of location. The kit contains 23 exploits across five full chains covering iOS 13.0 through 17.2.1, and Google says how it moved between these actors is unclear.

DarkSword followed a similar path. Commercial surveillance vendors and suspected state-sponsored actors have used it since at least November 2025 against targets in Saudi Arabia, Turkey, Malaysia, and Ukraine. The chain covers iOS 18.4 through 18.7 and runs every stage in JavaScript. Google says all six of its vulnerabilities are patched as of iOS 26.3, though most were fixed earlier.

How the Optimizely Attack Delivered Coruna and DarkSword to iPhone Visitors

Optimizely's status page says an unauthorized party exploited a vulnerability in Experimentation to add code to some customer projects, while Report URI wrote that it does not know how the attacker obtained write access to so many unrelated projects. It listed a compromised Optimizely system or a compromised credential with cross-customer access as possibilities, and said Optimizely's CDN served exactly what it was given, so this was not CDN cache poisoning. Report URI found that the loader sat in each project's Project JavaScript field, so the snippets still came from cdn.optimizely.com with Optimizely's own certificate, and the earliest injection it saw was in the BBC's snippet at 04:21 UTC on November 16. A mass cleanup began around 20:40 UTC, and by 23:11 UTC Report URI observed that all infected files had been cleaned.

Alongside a crypto address swapper, a card skimmer, and cookie and screenshot theft, the payload included an iOS branch that read the visitor's iOS version and opened an exploit chain in a hidden iframe, using Coruna for iOS 13 through 17, DarkSword for iOS 18, and, on iOS 26, a separate chain that so far only collects details about the device. Report URI said the Coruna loader was byte-identical to a sample from a different campaign it analyzed in October.

Cynet retrieved the second stage through an emulated iPhone browser, which shows conditional delivery rather than a compromise of a physical device, and Report URI notes that the kits' own comments mark iOS 18.7.3 and later, and iOS 26.3 and later, as patched.

Report URI states that it has no way to see how many iOS devices were infected before the attack was stopped. Its detection works through changes to trusted scripts, new script and iframe dependencies, and suspicious network destinations, which are signals from the website and its network connections rather than from the phone. The missing reports of compromise therefore reflect an evidence limitation and do not show that exploitation failed. What researchers confirmed is delivery of malicious JavaScript containing iOS exploit-kit loading functionality through compromised Optimizely integrations, which created a potential mass-exposure vector for previously documented exploit chains, while how many real iPhones received exploit payloads and whether any exploitation succeeded remain unestablished.

The kits themselves are not hypothetical, because Google observed three malware families deployed after successful DarkSword compromises in earlier campaigns and recovered the implant that Coruna delivers, so the capability behind the routing was demonstrated even though its success here was not. Which payload an Optimizely-routed iPhone would have received is unknown, since Coruna's final stage in the scam-site campaigns Google analyzed focused on financial data and cryptocurrency wallets, while DarkSword has delivered different payloads depending on the actor.

In the campaigns Google documented, victims reached the exploits through fake sites or compromised sites chosen for a target audience. Here the loader sat inside a script that CNN, the BBC, Zoom, and Salesforce already trusted, so reaching the kits required only that a visitor load a familiar page on an exploitable iOS version.

How Many iPhones Were Potentially Exposed to the Optimizely Attack? Are You At Risk? 

We have not found a reliable count. Public adoption statistics suggest that legacy iOS installations may number in the tens of millions or more worldwide, but adoption trackers report different version shares, and the number running specific exploitable versions has not been established. Exposure also does not equal compromise, because the devices that are vulnerable, the devices that loaded the malicious JavaScript, and the devices that were exploited are separate measurements, and only the first can be approximated from public data.

The sites on Report URI's list were ordinary destinations such as banks, retailers, streaming services, news outlets, and airlines, so a visitor had no reason to be suspicious and nothing required a click on a strange link. Our inference, which no published research establishes, is that people running unpatched iPhones skew toward those least able to notice a problem: an older relative who postpones updates, a teenager on a hand-me-down phone, a less technical user, or someone who cannot afford to replace a device that no longer gets updates.

The only visible symptom Report URI named was a page reloading every 60 seconds on iOS 18.6 or earlier, which most people would read as a glitchy site. Microphone capture raises the stakes for a younger family member, although we have no evidence that any Optimizely-routed device received a payload with that capability. The response therefore has to come from someone else in the household, such as the more technical person checking iOS versions and turning on automatic updates.

Could Coruna and DarkSword Enable iPhone Microphone Recording?

Google describes GHOSTKNIFE, a JavaScript backdoor that UNC6748 deployed after DarkSword compromises in Saudi Arabia, as collecting messages, accounts, browser data, and location history and recording microphone audio. Separately, Meta Red Team X published CVE-2026-65339 on November 9, a since-fixed flaw in Apple's audiomxd daemon that let an installed app record after the user declined permission and, on the earlier build Meta analyzed, without the orange indicator. It needed an app already on the device, there are no reports of in-the-wild use, and we have no evidence DarkSword used it, but it shows the microphone indicator is software that can fail too.

How iPhone Users and Website Owners Should Respond to the Optimizely Attack

Payloads that can collect messages, account details, and microphone audio make the exposure serious for anyone on an unpatched iPhone, even though success through this incident is unproven. Google's advice for both kits is to update to the latest iOS release, and to enable Lockdown Mode where an update is not possible. Report URI told anyone who visited an affected site on November 16 to be on iOS 18.7.3 or later, or iOS 26.3 or later, to check any crypto transfer made that day because a pasted address may have been swapped, and to watch card statements after paying on an affected site.

No single party owns remediation here, so the burden falls on individuals. Optimizely's status update says it removed the injected code, fixed the vulnerability, and secured affected accounts, but it gives no count of affected customers and does not address what happened to visitors. Report URI cannot see which visitors were exploited, so no public record shows who was affected. Updating iOS depends on the device owner installing the release, and we found no public sign that anyone is notifying individual visitors. Report URI advised site owners only to consider whether to tell their customers, which leaves anyone who typed a card number or pasted a crypto address on an affected site on November 16 to check their own accounts.

Site owners using Optimizely Web Experimentation should open the Project JavaScript in every project and the custom code in every experiment, look for anything they did not write, and then review which users, API tokens, and integrations can write to the account and rotate their credentials. Report URI advises treating cards entered in a site's own forms on November 16 as potentially exposed and rotating session tokens in cookies not marked HttpOnly. It recommends a Content Security Policy that restricts connect-src, avoids 'unsafe-eval' where possible, and turns on reporting, and it notes that allowing cdn.optimizely.com does not help because the injected code was served from that host.

The same trust relationship, a legitimate dependency carrying attacker code to everyone who loads it, applies to the MCP servers and Skills that AI agents pull in at runtime, as in our earlier write-up of SmartLoader cloning an Oura Ring MCP server to deploy a supply chain attack. 

If you are working out how third-party code reaches your own agents, talk with the Straiker team.

Sources and references

  1. Coruna: The Mysterious Journey of a Powerful iOS Exploit Kit (Google Threat Intelligence Group, March 3, 2026)
  2. The Proliferation of DarkSword: iOS Exploit Chain Adopted by Multiple Threat Actors (Google Threat Intelligence Group, March 18, 2026)
  3. World's Largest Brands Infected With Malware Via Optimizely (Scott Helme, Report URI, November 16, 2026, updated November 17)
  4. Hijacked Optimizely Scripts Served Card Skimmer and iOS Exploit Kits to CNN, BBC and Zoom Visitors (The Hacker News, November 17, 2026)
  5. Web Experimentation Security Incident (Optimizely status page, posted November 17, 2026)
  6. Who needs permissions and the orange dot that never lit (Meta Red Team X, November 9, 2026)

Key takeaways

  1. Report URI found malicious JavaScript in 188 Optimizely snippets belonging to 178 organizations on November 16, 2026, routing iPhone visitors toward the Coruna and DarkSword iOS exploit kits.
  2. Researchers confirmed malicious script delivery and exploit-kit loading functionality, but successful iPhone compromises and the number of affected devices remain unverified.
  3. Coruna and DarkSword have both appeared in campaigns involving surveillance vendors and suspected state-sponsored actors, with Coruna also reaching financially motivated attackers.
  4. The exploit kits target older iOS versions, while their own code comments identify iOS 18.7.3 and iOS 26.3 as patched release thresholds.
  5. iPhone users should update to the latest iOS version, enable Lockdown Mode if they cannot update, and review cryptocurrency transactions and card statements if they visited an affected site.

Ready to secure your own agents?

Talk to our team now
/ FAQ /

Frequently Asked Questions

Was my iPhone compromised if I visited a site affected by the Optimizely attack?

Compromise has not been publicly verified, but Report URI says it cannot see how many iOS devices were infected, so the absence of reports is not proof that no iPhone was exploited. Report URI says the kits' own comments mark iOS 18.7.3 and later and iOS 26.3 and later as patched, which makes an iPhone on an older release that loaded an affected page on November 16 the case worth checking.

Which iOS versions are vulnerable to Coruna and DarkSword?

The Optimizely attack used different exploit chains depending on the visitor's iOS version. The Coruna loader targeted iOS 13 through 17, while DarkSword targeted iOS 18. Report URI says comments in the exploit kits identify iOS 18.7.3 and later and iOS 26.3 and later as patched. The separate iOS 26 branch observed in the Optimizely payload collected device information rather than loading either exploit kit. Users should install the latest available iOS update rather than rely on a particular version threshold.

Does Lockdown Mode protect against Coruna and DarkSword?

Google recommends Lockdown Mode for devices that cannot be updated, and its Coruna analysis says the kit bails out when it detects Lockdown Mode. Google's DarkSword write-up gives the same fallback advice without saying how that chain behaves under Lockdown Mode, so updating iOS remains Google's primary recommendation.

Who was behind the Optimizely attack?

That has not been established in the coverage we reviewed. Report URI says it does not know how the attacker gained write access, and we found no public attribution tying the injection to the actors Google associates with Coruna or DarkSword. Report URI did tie the attacker's cash-out address to earlier operations in March and on November 3, 2026.

What else did the malicious Optimizely code do besides target iPhones?

It swapped cryptocurrency wallet addresses for the attacker's, skimmed cards typed into a site's own form fields (not hosted iframes such as Stripe), and sent non-HttpOnly cookies and page screenshots to the attacker. Report URI counted about $7,040 in the attacker's wallets by 19:25 UTC on November 16, and it cannot see how many cards, cookies, or screenshots were taken.

Does rebooting an iPhone remove these implants?

Published analyses do not document persistence across reboots, so a restart would ordinarily end a nonpersistent implant, though researchers have not shown that holds for every variant. Google notes that GHOSTKNIFE writes its working files under /tmp, which fits that picture. A reboot does not undo data already stolen, and an unpatched phone can be reinfected.

The Straiker agentic security platform

Secure AI agents across the full lifecycle

Discover AI

Visibility & Governance

Ascend AI

AI Adversarial Testing

Defend AI

Agentic Runtime Security

AGENTIC KILL SWITCH

// Secure with Straiker

Join the Frontlines of Agentic Security

You're building and using AI agents because the business demands it. Straiker gives your security team the visibility, testing, and runtime protection to keep up, without becoming a blocker. Deploy fast. Stay secure.