A compromised feature inside Optimizely's A/B testing platform turned websites operated by at least 178 organisations — including the BBC, CNN, HSBC and TalkTalk — into malware distributors. For UK MSPs and hosts, the real lesson isn't the headline victims; it's how invisible a marketing tool's supply chain can be to security teams.
View the data behind this chart
| Total scripts examined | Malicious scripts confirmed | |
|---|---|---|
| Optimizely scripts (17 Nov) | 1611 scripts | 188 scripts |
What actually happened on Optimizely's platform
Security researcher Scott Helme of Report URI identified malicious JavaScript being served from the Optimizely content delivery network to visitors of at least 178 organisations, including the BBC, CNN, Zoom, IKEA, Salesforce, HSBC and TalkTalk. The affected sites shared use of Optimizely Web Experimentation, a tool that lets businesses tweak page content and layout without touching the underlying site code.
Optimizely confirmed attackers exploited a vulnerability in its sign-up process, dubbed OPTIMJACK by researchers, to inject malicious code into customer projects through the feature that lets Optimizely users add their own JavaScript. Because the malware rode alongside Optimizely's own legitimate scripts from its CDN, customers had little visual reason to suspect anything was wrong.
Of 1,611 Optimizely scripts examined by Helme's team, 188 contained malicious code — including Optimizely scripts loaded by its own website and login portal. The payload evolved through the day: it began by silently swapping copied cryptocurrency wallet addresses for the attackers' own, then added a card skimmer that harvested numbers, expiry dates and security codes from payment forms. StepSecurity's analysis found that card data entered into an iframe hosted by a separate payment processor was out of the malware's reach — a detail with real architectural implications, covered below.
Why this is a supply-chain problem for MSPs, not just Optimizely's
A single flaw in Optimizely's sign-up flow allowed attackers to inject malicious JavaScript into customer projects, which was then delivered to visitors of affected sites through Optimizely's CDN alongside its own trusted scripts. That is precisely the pattern NCSC guidance warns about when it lists third-party software providers and website builders as established supply-chain attack paths.
For MSPs and hosting providers, the uncomfortable part is that experimentation, analytics and tag-management tools are usually chosen and configured by a client's marketing team, not vetted through the same change-control process as server infrastructure. This incident shows why asset inventories should include client-side third-party services.
Detecting a compromised experimentation script
The practical starting point is knowing exactly where Optimizely — or any equivalent experimentation platform — is loaded across a client estate, down to the script URL, version and any customer-added JavaScript. Without that inventory, there is no baseline to compare against when a vendor is compromised.
- •Enumerate every site and tenant loading Optimizely (or similar) scripts, recording script URLs, hashes and versions
- •Pull historical responses for those scripts and diff them against what is being served now
- •Check browser and CDN logs for unexpected script changes, clipboard access, form-field reads or wallet-address substitution
- •Flag outbound requests to unfamiliar domains that coincide with page loads of the affected script

Containment and credential rotation once compromise is suspected
Speed of containment matters more than certainty. Where business impact allows, disable the Optimizely integration and use Content Security Policy to restrict script sources and outbound connections, rather than waiting for full confirmation; CSP alone may not block malicious code served from an already-trusted origin unless the integration is removed or constrained.
As a precaution, treat credentials associated with affected projects as potentially exposed and rotate them where appropriate. NCSC's dependency guidance recommends rotating exposed credentials, enforcing MFA, auditing for unauthorised token use, and — as a risk-based precaution scoped to affected integrations and change processes — pausing automatic dependency updates while remediation is confirmed. This is the point where incident response planning either pays for itself or doesn't exist at all.
Running the client notification playbook
Every client whose site loaded the affected service needs to be notified, including those where compromise isn't yet confirmed. Silence is the worse option — the attackers' payload changed multiple times in a single day, so organisations needed to recheck after remediation. By 2152 UTC on the Monday of the attack, TalkTalk was the last organisation still showing the malicious code on Report URI's tracker, showing that malicious code remained observable on at least one customer site after the vendor said it had fixed the vulnerability.
A credible notification separates confirmed exposure, probable exposure and assessed non-exposure, and gives each tier the detail needed to act: affected domains and time windows, the observed payload behaviour, containment steps already taken, and what evidence has been preserved. Ask customers to review payment pages, wallet transactions, CSP violation reports, WAF events and endpoint alerts across the exposure window, and to replace credentials or payment instruments where there is evidence of collection.
View the data behind this chart
| No inventory | Basic monitoring | Full playbook | |
|---|---|---|---|
| Detection time | Days (customer report) | Hours (log review) | Minutes (SRI/CSP alert) |
| Credential rotation | Not performed | Manual, delayed | Automated, immediate |
| Client notification | Ad hoc or none | Generic advisory | Tiered, evidence-based |
| Contractual assurance | None | Informal | NCSC-aligned SLA clauses |
Architectural lessons worth acting on now
This is not the first time a trusted third-party script has turned visitors' browsers against them; NCSC's Browsealoud guidance documents the same failure mode and recommends risk-based decisions about which third-party scripts a site really needs, local hosting where practical, Subresource Integrity, CSP, change monitoring and a rapid takedown process.
SRI may be impractical for frequently changing experimentation assets; assess whether stable loader scripts can be pinned. The more durable fix is scoping CSP allow-lists tightly and running a supplier cyber risk assessment on every analytics, tag-management and support-chat vendor with write access to production pages. StepSecurity found that card details entered into the analysed third-party payment iframe were outside this malware's reach; architecture should be assessed case by case. Contractually, NCSC also recommends requiring key suppliers to report security performance, accept audit rights, and flow equivalent requirements through their own subcontractors — terms worth revisiting at the next renewal. A free check with our security headers checker can help review your CSP configuration.
- 01The Register — Optimizely hack turns world's biggest websites into malware dispensers · 17 November 2026
- 02NCSC — Software supply chain attacks: check your dependencies · 1 July 2026
- 03NCSC — Supply chain security: understand the risks · 1 January 2026
- 04NCSC — Supply chain security: check your arrangements · 1 January 2026
- 05NCSC — Malicious software used to illegally mine cryptocurrency · 1 January 2026
- 06ncsc.gov.uk
- 07ncsc.gov.uk
- 08ncsc.gov.uk
