All articles

Experimentation security

After the Optimizely hack: move your section tests to Patchwork

What Optimizely confirmed, why testing scripts deserve scrutiny, and how to move section experiments to Patchwork with stricter content boundaries.

Incident information checked November 17, 2026, 03:31 UTC.

What Optimizely has confirmed

Optimizely's Web Experimentation security notice says an unauthorized party exploited a vulnerability and added malicious JavaScript to some customer projects. In its November 17 update at 01:02 UTC, Optimizely said it had fixed the vulnerability, removed the injected code, secured affected accounts and added protections against unauthorized code delivery. The incident was listed as monitoring when we checked.

That is the confirmed vendor account. We are not treating every Optimizely customer as affected, attributing the attack to a particular group, or assuming the incident remains uncontained. Follow the official notice for updates relevant to your own account.

Your testing script is part of your security boundary

A tool that changes a live page is a production dependency. Its permissions, publishing controls and delivered code deserve the same attention as other changes to your site. Replacing a vendor does not, by itself, remove supply-chain risk.

For teams whose job is testing a hero, pricing section or call to action, our recommendation is to move to Patchwork's narrower section-testing workflow. You get AI-designed variants, testing and rollout for $100/month, with explicit limits on what variant content can contain.

Choose concrete controls over a bigger security promise

Patchwork's variant pipeline sanitizes HTML and scopes CSS to a variant wrapper. It removes script elements, event handlers, iframes and unsafe URL schemes. The CSS pipeline rejects imports and unsafe values. These are implemented content restrictions, not an assertion that Patchwork cannot be compromised.

AI-generated variants wait for approval. Experiment edits require an owner or editor role; viewers cannot publish changes. Section rendering has a timeout that restores the original when a variant cannot render. A rendering fallback protects availability; it is not a malware detector.

Optional Optimizely-compatible callback and polling activation can run a reviewed function on your site after consent. That code has your site's browser privileges and needs CSP permission for Function compilation. Keep manual activation if you require a policy that disallows eval. HTML/CSS sanitization does not sandbox a saved activation function.

URL forcing is disabled by default, and draft/paused content is excluded. An owner or editor can explicitly enable public disabled previews through the compatibility API. That setting delivers unapproved content publicly; PUBLIC is a preview switch, not authentication. Those previews stay outside ordinary assignment and measurement, and disabling them cannot recall content already downloaded.

Migrate the experiment, not suspicious executable code

Start from a known-good version of your page and rebuild the section test. Do not paste an unreviewed project script into a new vendor's custom-code field. Keep your old results as historical records and collect a fresh baseline after changing assignment or measurement.

Patchwork's migration tool reads Optimizely's documented v2 API and experiment exports, identifies unsupported behavior and moves supported section changes into review drafts. The migration boundary is deliberate: executable changes, advanced audiences and other unsupported settings are called out, rather than silently carried into a new test.

  • Confirm your exposure using Optimizely's incident guidance and your own script inventory and access logs.
  • Review publishing access, API credentials and delivered script changes before resuming affected tests.
  • Capture a known-good page section, review every imported variant and configure a new conversion goal or validate the mapped goals and custom-event instrumentation.
  • Validate the new installation and stop the old experiment before sending production traffic to a replacement.

Is Patchwork a drop-in replacement for Optimizely?

Not for Optimizely's entire platform or browser SDK. Patchwork's native section-testing workflow uses its own snippet, API keys and UUIDs. Its opt-in compatibility API preserves bound source IDs for documented project, experiment, page and event operations, including prelaunch metrics and linked native conversion goals. Backend feature flags, advanced audience logic, historical assignments and arbitrary project JavaScript are not interchangeable.

A credible drop-in claim needs documented compatibility and tests for the exact API or SDK surface being replaced. That is the standard we are building toward. Today, choose Patchwork for its focused workflow and verifiable content restrictions, rather than an untested promise of full parity or greater overall security.

Sources

Product-control descriptions were checked against Patchwork's implementation. No independent comparative security audit is claimed. Incident details can change; consult the linked vendor notice.