Hudson Rock found malicious code inside Optimizely projectJS configurations served to customers on major websites. The recovered files included a minute-ly loader and poisoned third-party JavaScript linked to browser-based financial theft and delivery of iPhone exploit-kit stages.
That left 21 named web properties tied to the evidence, including, Zoom and GitLab. Kirk, a research contributor from Ohey, provided archived response verification and iOS-routing analysis that Hudson Rock folded into the report.
Optimizely files and vendor endpoints
The malicious code sat inside customer projectJS configurations alongside legitimate experimentation code in Optimizely. Some Optimizely versions used a plain loader, while others encoded the URL, and the affected files were served from third-party providers' own script and widget endpoints.
Hudson Rock directly recovered nine poisoned vendor files and identified eight distinct recovered Optimizely assets. That matters operationally because the evidence points to poisoned delivery files rather than a generic backend break-in, which narrows the likely attack surface without proving where the original insertion began.
, Zoom and GitLab evidence
Kirk's 15-capture table overlapped Hudson Rock's findings for, Zoom and GitLab, and it added 12 properties. The same broader evidence set also covered, Salesforce, Mutual of Omaha and, while Talkwalker's homepage HTML directly referenced 4894383279964160.js and The file was /js/5728063363743744.js.
The and HSBC UAE specimens embedded a complete fallback payload, and Hootsuite and Talkwalker shared one poisoned project. Hudson Rock thanks Kirk from Ohey for bringing this incident to Hudson Rock’s research team’s attention and for providing the additional archived response verification and iOS-routing analysis incorporated into this report.
What the evidence does not prove
The source material frames this as a documented subset of activity, not an incident-wide victim count. It does not establish how many malicious page loads succeeded, how many devices were compromised, or whether every visitor loaded the malicious version.
The same evidence also does not establish a compromise of the named websites' backend systems, or whether provider infrastructure, publishing credentials, integrations or separate customer accounts were involved. Kirk reported 51 traced sites in the wider investigation, but the visible record here still leaves the insertion path open.
For affected teams, the practical next step is to treat the poisoned projectJS delivery path as the problem to isolate, not just the visible page code. The unresolved question is how the malicious Optimizely code was introduced into customer projectJS configurations.







