A recognizable website and a valid HTTPS certificate do not guarantee that every script running in a visitor's browser is trustworthy. On 16 November 2026, a campaign known as OPTIMJACK used malicious JavaScript delivered through Optimizely Web Experimentation scripts on legitimate websites across media, software, financial services and crypto-related businesses.
In his 16 November investigation, Scott Helme of Report URI reported 188 infected snippets among 1,611 examined, spanning 178 organizations. Those are counts of scripts and organizations identified in the research sample. They do not establish that 178 companies' own servers were breached or that 178 confirmed cases of data theft occurred.
On 17 November, Optimizely officially confirmed the security incident. The company said an unauthorized party had exploited a vulnerability in Experimentation to inject malicious JavaScript into customer projects. It reported fixing the vulnerability, removing injected code and securing affected accounts. The latest official update reviewed on 17 November classified the incident as Monitoring.
How a trusted third-party script became the attack path
Websites use Optimizely to run tests and personalize experiences. A visitor's browser may load project-specific JavaScript from cdn.optimizely.com. These files can contain custom code from the customer's Project JavaScript field.
Report URI identified malicious loaders in that project code. The files still arrived through the expected CDN and HTTPS connection: the content had been modified upstream, not tampered with while traveling over the network. This does not in itself imply compromise of every affected website's origin servers.
Researchers documented different loader variants: an explicit attacker URL, a base64-encoded reference and a version carrying an inline copy of the malicious payload. A clean current copy of a script, or a text search for only one malicious domain, cannot prove the script was clean on 16 November.
StepSecurity's analysis published on 17 November correlated public web scans, network traces and script variants. The underlying trust failure is the same: code from an approved vendor may operate with the privileges granted to the page's JavaScript.
Four capabilities, and why the level of proof matters
1. Cryptocurrency address replacement
The crypto clipper component searches for addresses used near payment or transfer-related text and may replace them with attacker-controlled addresses. Researchers identified targets on Bitcoin, Ethereum and other EVM-compatible networks, Solana, TRON, TON, Litecoin, Bitcoin Cash and Dogecoin.
The analyzed behavior includes replacement of displayed text, selected inputs and links, copy events and certain JavaScript requests. Consequently, an address pasted into a wallet application might not be the destination the user intended. A syntactically valid address and an otherwise valid signed transaction do not demonstrate recipient authenticity.
A hardware wallet cannot infer the user's intended recipient. If its screen displays an attacker address and the user confirms it without independent verification, the transaction can still be authorized. Compare the complete destination, relevant contract and amount on the signing device against a destination confirmed through a channel independent of the potentially compromised website—not just another element of that same webpage.
2. Screenshots and JavaScript-readable cookies
The analyzed payload can capture a visible part of the page and cookies accessible to JavaScript once certain triggering conditions occur, including an address replacement. HttpOnly cookies are not readable through this mechanism.
A capability to collect data, an observed exfiltration request and proven harm to a particular visitor are three different propositions. Not every page view would trigger the same behavior, and the volume of information actually received by the attackers is not publicly established.
3. Payment-card skimming attempts
A separate component watches accessible forms and network calls for card numbers, expiration dates and verification codes that pass formatting checks. It can attempt to send this information to an attacker-operated endpoint.
Card fields isolated inside a cross-origin payment iframe were not directly readable by this script according to StepSecurity, because the browser's same-origin policy separates them. That does not mean every payment integration is safe from all attacks; the actual page and processor design matter.
4. iOS-specific exploit delivery attempts
The campaign also referenced additional components aimed at iPhone and iPad visitors. Report URI associated these with Coruna and DarkSword exploit chains and described version-based targeting.
Important distinction: loading or attempting to load an exploit kit does not prove successful compromise of a device. StepSecurity did not independently reproduce every iOS stage, and there is no reliable public count of successfully exploited devices. Users potentially exposed should install the relevant Apple security updates while treating infection claims as unproven without device-specific evidence.
Timeline and impact: separating evidence from inference
| Item | Documented observation | What it does not prove |
|---|---|---|
| 16 November, from around 04:21 UTC | First Optimizely injection seen by Report URI in the captures studied | The attacker's full campaign may have started earlier |
| 1,611 snippets examined | 188 infected snippets associated with 178 organizations | Not a count of individual visitors or confirmed victims |
| 16 November, evening UTC | Mass cleanup; one snippet still infected at 21:52 UTC in Report URI's checks | Clean content today does not establish prior safety |
| 17 November | Optimizely confirms the vulnerability and reports fixes and cleanup | Does not identify all downstream data theft |
| About US$7,040 | Funds received at addresses tracked by Report URI as of 19:25 UTC on 16 November | Some wallet activity may predate this particular operation |
Research mentions affected scripts loaded by pages associated with BBC, CNN, Zoom, Salesforce, GitLab and Ledger, among others. A malicious snippet on a page associated with Ledger does not establish compromise of Ledger hardware wallets. The documented browser risk is separate from the integrity of a physical signing device.
Six checks for website owners and defenders
- Inventory every Optimizely Web project included across websites and subdomains, including scripts served from first-party domains or loaded by a consent manager.
- Review Project JavaScript content and change history, account access, API tokens and unusual administrative changes. Ask Optimizely for affected project identifiers and timestamps.
- Search historical telemetry for indicators, including 1688360[.]cc, webshell[.]lol, relevant .icu domains, and paths such as /track.js, /hit.php and /card.php. Also examine encoded and inline variants and correlate browser, proxy and CSP report data.
- Preserve evidence before further changes: historic JavaScript responses, network captures, timestamps, hashes, project settings and incident-handling decisions. Public sources such as urlscan.io may provide clues but have incomplete coverage.
- Reduce third-party exposure on sensitive pages. Where practical, remove nonessential tags from payment and authentication flows, favor isolated cross-origin payment fields and review vendor access.
- Test Content Security Policy (CSP) in practice, especially script-src and connect-src, checking which unexpected loads and destinations are blocked. Proper CSP can reduce the scope of some attacks, but strict-dynamic can extend trust to scripts loaded by already trusted scripts. A domain allowlist alone is not a universal defense.
The MDN CSP guide explains the controls. Subresource Integrity (SRI) can verify a known static file hash, but pinning the hash of a frequently changing experimentation snippet can be operationally difficult. Combine technical controls, historical monitoring and vendor governance rather than relying on one mechanism.
Initial investigation window: at least 16 November 2026 from about 04:00 UTC into that evening. Adjust it for the site's exposure and any later details from the vendor. A blocked CSP request is a useful indicator, not proof that a theft succeeded.
What should potentially exposed visitors do?
Someone who made a cryptocurrency transfer during the relevant window should retrieve the intended destination through an independent trusted channel, then compare it to the address in the signed and published transaction. Never disclose a recovery phrase in order to “verify” a transaction. If funds went to the wrong address, preserve transaction evidence and promptly seek appropriate support without assuming that recovery will be possible.
Someone who entered payment-card information into an affected accessible form should contact their bank about the possible exposure and monitor transactions; follow the institution's specific advice. iPhone and iPad users should apply available Apple security updates. Organizations may consider targeted, authorized forensic triage when exposure and other indicators justify it.
What remains unknown
The complete technical root cause and how the vulnerability may have allowed changes across many unrelated customer projects are not yet fully established publicly. There is no definitive public attacker attribution, consolidated count of stolen payment cards or confirmed number of exploited iOS devices. Nor should an unverified CVE identifier be invented for this incident.
The broader lesson extends beyond cryptocurrency: organizations that execute third-party JavaScript inherit some of their suppliers' operational risk.
GLOV Consulting covers organizational decision-making and risk management, while the principles described by GLOV Secure apply to safeguarding crypto assets and verifying transactions. Our earlier reports on the malicious Rank Math PRO update and the LiteLLM supply-chain compromise illustrate different failure modes involving trusted providers.
Information reviewed 17 November 2026. This article is educational; it does not allege any compromise of GLOV Solutions or its customers. Digital forensics and GLOV AI are still shown as upcoming activities on the website.
Official sources and independent technical analyses
- Optimizely — official Web Experimentation security incident, 17 November 2026.
- Scott Helme / Report URI — original research, scope, timeline and organizations (16 November 2026).
- StepSecurity — OPTIMJACK loader variants, crypto clipping, card skimming and IoCs (17 November 2026).
- BleepingComputer — incident reporting and independent context (16 November 2026).
- MDN — Content Security Policy guide.
- MDN — Subresource Integrity guide.