Threat Landscape

OPTIMJACK’s real lesson: the vendor’s install guide punched the hole in your browser defenses

When attackers poisoned Optimizely snippets on 178 organizations’ sites, Content Security Policy fully stopped the payload at only 24 of the 159 Report URI checked. HSBC blocked the main script and still got hit, because Optimizely’s docs tell customers to allow eval. Every third-party script now writes part of your security policy.

Nov 17, 2026 · 4 min read

The short version. On November 16, an attacker wrote a malicious loader into the Optimizely Web snippets of 178 organizations, including Zoom, CNN, the BBC, Salesforce, GitLab, HSBC and Equifax. We covered the breach itself last night. This piece is about something else: why so many well-run security programs had a defense in place and it still didn’t work.

The mechanics in one paragraph. Optimizely Web is an A/B-testing tool. Each customer site loads one JavaScript file, the snippet, from cdn.optimizely.com on every page. The snippet has a field called Project JavaScript where customers can add their own code. Optimizely says an unauthorized party exploited a vulnerability in its Experimentation account signup flow to add malicious code to customer projects. It says it fixed the flaw, removed the code, ended sessions, revoked tokens and blocked delivery at Cloudflare. It has not said how a signup bug gave write access to other companies’ projects. Report URI, Scott Helme’s company, found the loader in 188 of the 1,611 snippets it checked. The payload skimmed cards typed into a site’s own forms, swapped crypto wallet addresses, stole readable cookies and screenshots, and pushed iPhone exploit kits.

The number nobody is talking about. Report URI crawled the affected organizations’ own websites and tested each site’s Content Security Policy against what the payload needed. CSP is a browser rule set a site publishes that says which domains may run scripts and where the page may send data. It is the textbook defense against exactly this kind of attack. Here is how it went. 68 organizations had no CSP at all. 50 had a CSP that allowed both the payload and the data theft. 17 let the payload run but blocked it from sending data out, though the wallet swapping still worked. Only 24, about 15%, blocked it outright.

That is the failure. Not “nobody had controls.” Most of the sites that had controls had configured them into uselessness.

How the policy got loosened. Two settings did most of the damage, per Report URI. The first is 'strict-dynamic', a modern CSP option that says: if I trust a script, I trust any script it loads. The poisoned snippet came from a trusted host, so the attacker’s child script inherited that trust. The second is 'unsafe-eval', which lets code turn text into running code. Report URI says Optimizely’s own documentation tells customers to allow it. One variant of the attack carried its full payload inline and ran it through eval when the attacker’s server was blocked. That is what happened at HSBC: its CSP blocked the main malicious script and the data theft, and the inline copy ran anyway.

So the hole wasn’t a mistake by the security team. It was a setting the vendor needed. Every third-party script on your page quietly negotiates your security policy down to what that vendor requires. Multiply that by the analytics, consent, chat and testing tags on a typical checkout page.

This was already a compliance requirement. Since March 31, 2025, PCI DSS v4.0.1 requirements 6.4.3 and 11.6.1 have required merchants to authorize and inventory every script on payment pages, check each one’s integrity, and detect unauthorized changes to scripts and security headers as the browser receives them. OPTIMJACK was a live audit of those two controls. Report URI’s sector table puts 32 of the 178 organizations in banking and finance. Expect assessors and card brands to ask pointed questions this quarter.

The CyberMerge view. Third-party risk programs ask vendors about SOC 2 reports and breach history. Almost none ask the question that mattered here: what does your script need us to switch off? That belongs in procurement, next to the security questionnaire.

Three asks for buyers. First, make every script vendor document the CSP exceptions it needs, and refuse 'unsafe-eval' on login and payment pages. Second, lock down where your pages can send data (CSP connect-src). That alone protected 17 sites from the card and cookie theft. Third, monitor what third-party scripts actually contain from day to day, because a URL that never changes can serve different code tomorrow. Report URI kept dated copies of every infected file. Most companies could not tell you what their snippet said on Sunday.

For vendors, the signup-flow detail is the scarier one. If a self-service signup bug can reach other tenants’ projects, that is a multi-tenant isolation failure, not a script problem. Optimizely owes customers that root cause.

Bottom line. The browser is now a supply chain, and the vendor’s install guide is part of your attack surface. The sites that came through clean didn’t buy a new product. They kept a tight policy even when a vendor asked them to loosen it.

Sources