Optimizely Script Injection Exposed Visitors to Crypto Address Swaps and Card Skimming

In reporting first published on November 17, 2026, security researchers described a malicious JavaScript campaign affecting website scripts used by 178 organizations, including Ledger, OANDA/Coinpass, IG, Interactive Investor and TMS Brokers. Other affected brands included the BBC, Salesforce and Zoom. The attack could replace a cryptocurrency address copied by a visitor with an attacker-controlled address; it also targeted card details entered into certain website forms.

The practical risk: A visitor who copied a wallet address from an affected page could paste the attacker’s address into a transfer without noticing the change. Check the full destination address after pasting it and before confirming a transfer. The reporting does not establish that every visitor to an affected site encountered the malicious code or lost funds.

This does not mean each organization’s internal systems were breached. An OANDA spokesperson, speaking for Coinpass and TMS Brokers, said its internal systems and infrastructure were not breached and that it had seen no indication of an impact on client funds or account data. Salesforce said it was investigating unauthorized changes to Optimizely scripts on certain Salesforce websites and, as a precaution, was confirming those scripts were disabled across preview and production instances.

Optimizely Script Injection Exposed Visitors to Crypto Address Swaps and Card Skimming (1)

How the injection worked

Optimizely allows customers to run experiments and custom code through JavaScript snippets on their websites. An Optimizely spokesperson said an unauthorized party exploited a vulnerability in its Experimentation account signup flow to add malicious JavaScript to some customer projects. According to the spokesperson, Optimizely blocked the vulnerable path, deployed a fix, removed injected code, ended active sessions, revoked tokens and reset access for affected accounts. It also blocked the code from being delivered in Cloudflare, notified affected customers and continued its investigation.

Once an affected script loaded, the malicious code could substitute attacker-controlled crypto addresses, including when a visitor copied an address to the clipboard. StepSecurity’s analysis also found code that captured card numbers, expiry dates and security codes entered into website forms. It could not read payment fields isolated in a third-party provider’s cross-origin frame, StepSecurity said. Report URI CEO Scott Helme said the script also loaded code linked to the Coruna and DarkSword iOS exploit kits for visitors using iPhones and iPads. Researchers had not established how the attackers obtained access to modify the Optimizely projects.

According to Helme, the BBC was among the first sites observed serving the injection at 4:21 am UTC on Monday, November 16. By 9:52 pm UTC that day, one of 188 infected snippets still contained malicious code; Ledger’s snippet had been cleaned at about 8:41 pm UTC. The 188 figure counts snippets, while 178 counts organizations.

Optimizely Script Injection Exposed Visitors to Crypto Address Swaps and Card Skimming (2)

What is known about losses

The amounts traced to associated crypto addresses were not confirmed victim losses. Blockchain records checked by Cointelegraph showed around $7,100 received on Monday. StepSecurity separately estimated roughly $6,600 received or held across the wallets and noted that some addresses had activity before the attack. Galaxy Digital’s Alex Thorn estimated that victims sent around $2,300 in total; he said most Bitcoin receipts appeared to be the attacker’s own funds routed through an exchange and back to their wallet.

Helme said researchers had no evidence that payment-card details were actually captured. He said the card skimmer was added partway through Monday’s campaign and that the realistic risk was concentrated on sites collecting card numbers in their own forms during the exposure window, rather than sites using hosted payment frames. These findings describe what the code could do and what researchers observed—not a confirmed loss for every affected site or visitor.

Further reading