Optimizely Flaw Turned 178 Sites Into Malware Delivery

Introduction
For roughly 24 hours, visitors to some of the most heavily trafficked destinations on the web — the BBC, CNN, Equifax, Ledger and Zoom among them — were served malicious JavaScript that had nothing to do with those organizations. According to a blog post from security firm Report URI, at least 178 web properties unwittingly delivered malware through Optimizely, the experimentation platform many companies use to customize user experiences and trial new features. The injected code was rebuilt 26 times over the course of the day, letting whoever was behind it iterate on payloads in near real time.
The incident is a textbook supply chain compromise in which the victim organizations' own code and infrastructure were never breached. A vulnerability inside a vendor's platform did the work — and the failure of most sites to constrain which domains may execute script on their pages turned a single vendor flaw into a mass exploitation event.
What the attackers actually delivered
Report URI researcher Scott Helme documented that malicious JavaScript was served to visitors of 178 organizations through Optimizely's platform on November 16. The payload was not static: it was rebuilt 26 times during the day, dropping a multi-purpose malware kit capable of stealing payment card data, cookies and screenshots, and of swapping cryptocurrency payment addresses in the browser for ones controlled by the attacker. That last capability is especially dangerous for anyone who copies a wallet address from a web page and pastes it into a transfer form.
Optimizely confirmed the incident, stating that it identified a security incident in which an unauthorized party exploited a vulnerability in Optimizely Experimentation to add malicious JavaScript to some customer projects. The company said it blocked access to the vulnerable path, deployed a fix, removed injected code from affected projects, ended active sessions, revoked tokens and reset access for affected user accounts, and added protections intended to stop malicious code from being delivered in Cloudflare. It said the incident has been fully remediated and that third-party experts are assisting with a comprehensive investigation. The technical nature of the underlying vulnerability has not been publicly detailed.
Older iPhones and iPads in the crosshairs
Beyond the data-theft kit, the affected sites also served exploit code built specifically to attack known vulnerabilities in older Apple iPhone and iPad devices. Report URI named the exploit kits DarkSword and Coruna — zero-click chains that string together roughly half a dozen security holes to seize complete control of an outdated device and extract data from it. Zero-click is the critical word: the visitor did nothing wrong. No app install, no suspicious link, no permission prompt.
The targeting choice is telling. Exploit kits aimed at unpatched mobile devices monetize the long tail of hardware that no longer receives updates or whose owners defer them. For defenders, it is a reminder that client-side patching is part of the web attack surface, not a separate housekeeping task owned by someone else.
Why the blast radius was so wide
Modern web pages are assembled at runtime from many sources. According to the 2025 Web Almanac published by HTTP Archive, the median homepage loads at least 22 JavaScript files, and many top properties will not fully render content until visitors accept dozens of third parties serving script. JavaScript runs on the visitor's device and can read what the browser displays, alter page content and quietly ship data elsewhere.
Zach Edwards, a researcher at Infoblox and co-founder of the ad-tracking service DecryptAds, noted that Optimizely snippets are typically embedded as close to the top of the page as possible and are configured to fire immediately on page load rather than wait behind a consent dialog, because the product is built for A/B testing and needs to render early. The practical consequence is that this campaign required no click and no consent to trigger — most affected clients would have served the payload the moment someone loaded the page.
The CSP gap, in numbers
Content Security Policies let a site declare which domains are permitted to serve it script. Report URI's analysis of the affected properties found that 68 had no CSP at all, while 50 had a policy permissive enough to allow both the malware payload and the resulting data theft. Only 24 of the affected sites — about 15 percent — had policies that fully blocked the malicious payload.
Baseline adoption across the busiest sites is lower still. Helme's crawler.ninja, which scans top sites, puts overall CSP deployment at roughly 10 percent. Edwards observed that most organizations outside the largest e-commerce, social media and financial platforms either skip CSP entirely or leave it deliberately loose for marketing reasons, and that without a policy, a compromise at any embedded partner allows new domains to be injected into the page unchallenged.
Grivyonx Expert Analysis
The uncomfortable lesson here is that the attack path was a legitimate product feature. Optimizely exists to inject and modify code on customer pages at speed, without a deploy, and the attackers simply borrowed that capability. Any vendor with standing permission to rewrite your front end — tag managers, experimentation platforms, personalization engines, chat widgets — is effectively a remote code execution channel into your users' browsers, and should be governed as one. The numbers from Report URI show the governance gap clearly: a majority of the impacted sites had either no policy or a policy written for convenience rather than containment. CSP is not a silver bullet, but it is the only browser-native control that decides, at request time, whether a newly injected script is allowed to run. Teams that treat it as a marketing obstacle rather than a security boundary are choosing to outsource their incident response to a third party's patching speed.
What security teams should do now
The Optimizely vulnerability has reportedly been remediated, but the structural exposure it revealed has not changed. Practical steps:
- Inventory every third-party script on production pages, including those loaded by tag managers and experimentation tools, and record who owns each relationship and what the business justification is.
- Deploy CSP in report-only mode first, collect violation reports, then move to enforcement with nonces or hashes and strict-dynamic rather than broad domain allowlists that include entire marketing ecosystems.
- Alert on new script origins. A policy that permits a compromised partner to pull from an unseen domain is not a control. Monitor violation reports and treat unexpected origins as an incident signal.
- Review pre-consent execution. Ask vendors whether their snippet must fire before a visitor interacts, and push back where the answer is habit rather than necessity.
- Rotate credentials with script-injection rights. Accounts that can publish code to customer properties deserve the same session hygiene and token rotation as production deploy keys.
- Patch client devices. The DarkSword and Coruna chains depended on unpatched iOS and iPadOS versions; asset management has to include phones and tablets, not just servers.
Conclusion
Trusted JavaScript does not stay trustworthy by default, and a vendor's security posture is now part of your own attack surface. The Optimizely incident will be remembered less for the specific flaw than for how much of the web had no mechanism to say no when that flaw was exploited. Organizations that want to shrink this class of risk should combine vendor scrutiny with browser-enforced boundaries — and where internal teams are stretched, partners offering AI automation, DevSecOps and cloud security services can help make continuous script inventory and policy enforcement a routine part of the pipeline rather than a one-off cleanup.
- #Optimizely
- #supply chain attack
- #Content Security Policy
- #third-party JavaScript
- #malware
- #web security
- #Report URI



