A software supply chain attack delivered malware to visitors of 178 organizations’ websites on Nov 16. — without directly breaching the companies themselves. Attackers weaponized a marketing tool embedded across those sites, compromising some of the world’s largest brands and raising questions for enterprise CISOs about the trustworthiness of third-party scripts.
Web security vendor Report URI uncovered the threat campaign, in which attackers injected malicious JavaScript into Optimizely Web snippets — tools organizations use to create, test and optimize online content. The campaign included 188 infected snippets belonging to 178 organizations, including Salesforce, Zoom, GitLab, Equifax, HSBC, CNN and Ikea.
By 21:52 UTC, most of the malicious code had been removed, with the malicious script found in just 1 of the 188 infected snippets.
The malware arrived with a valid certificate from Optimizely’s own content delivery network, but how the attacker gained write access to unrelated customers’ Optimizely projects is unclear. TechTarget reached out to Optimizely representatives but, as of publication, has not heard back.
“The uncomfortable lesson is that none of the victims were breached,” Scott Helme, the security researcher who discovered the malware campaign, told TechTarget. “Their own infrastructure was untouched; a vendor’s script turned hostile and their sites served it to every visitor.”
According to Report URI, the payload swapped cryptocurrency wallet addresses, skimmed payment cards, stole cookies and took screenshots. On iPhones and iPads, it delivered the Coruna and DarkSword iOS exploit kits. According to Report URI, as of the evening of Nov. 16, the attacker had stolen about $7,000 in cryptocurrency from targeted organizations’ web users.
“But money is only the part we can count,” cautioned Peter Lowe, threat research lead at threat intelligence vendor ZeroFox. “Non-HttpOnly cookies, card details typed into sites’ own forms rather than a hosted payment frame, and the iOS exploit chains served to anyone on an unpatched iPhone are all impacts nobody can measure from the outside.”
Lowe advised organizations to audit their code, looking for any instances that incorporate JavaScript or other third-party assets, and then set up monitoring. Optimizely Web customers should “check the Project JavaScript yourself and rotate everything that can write to it — users, tokens, integrations,” he added.
Rethinking ‘trusted’ scripts
One of the incident’s major takeaways for CISOs is that enterprises cannot unconditionally trust third-party scripts.
“When it runs on your page and skims your customer’s card, it’s your breach, not your vendor’s,” said George Gerchow, IANS faculty member and CSO at data security firm Bedrock Data. He added that every company running third-party scripts should know exactly which scripts can execute code on their login and payment pages, lock down where that code can send data, and be alerted the minute a script changes.
“This one ran for about 17 hours on some of the biggest sites in the world,” Gerchow said.
Self-hosting isn’t necessarily more secure. According to Helme, some of the victims served Optimizely’s snippet from their own domains and were still infected because they proxied the live file.
“The practical starting point is a Content Security Policy (CSP) in report-only mode,” Lowe said. “It breaks nothing; it just has every visitor's browser tell you what loaded and where data went, and you end up with an inventory built from real traffic rather than a spreadsheet from 2023.”
According to Helme, most of the affected organizations lacked a CSP or monitoring tool capable of detecting the attack. He and Lowe suggested setting up alerts for any unexpected new scripts, unexpected changes to a trusted file, and data flowing to new destinations.
“CSP reporting can flag the kind of fetch and sendBeacon requests the skimmer used to exfiltrate data,” Helme said.
He added that CISOs should also push back on vendors that require ‘unsafe-eval’ or isolate them. Report URI reported that a CSP blocked the external payload on HSBC’s sites, but an inline copy embedded in the snippet could still run because the policy allowed eval.
Organizations should also inventory and minimize the use of scripts on payment and login pages, Helme advised.
“An A/B testing tool rarely needs to be on a checkout or account page at all,” he said. “‘Trusted’ should mean ‘allowed to do these specific things,’ not ‘allowed to do anything.’”
Ultimately, according to Lowe, the question isn’t whether an organization trusts its third parties — it’s whether it would notice if one of them quietly changed in the middle of the night.
“Speed is important here,” he said. “This incident ran from about 4 a.m. UTC until most sites were cleaned at around 9 p.m., so a report on Friday wouldn't have been much good.”
Similar future incidents will involve enterprise AI agents, predicted Gerchow.
“The plugin or MCP server you trusted yesterday is the delivery channel tomorrow, and an agent will run whatever it’s handed,” he warned. “The true test is whether you’d know when what your vendor serves you changes.”
Alissa Irei is a TechTarget senior reporter covering cybersecurity news.