Zero trust in the AI era: how to audit the third-party applications inside your software
AI is arriving in enterprise software faster than most governance processes can follow. Every new copilot, analytics tag, personalisation engine and plug-in is another supplier with a route into your web estate, your customer data, or both. At that pace, no brand is safe by default, whatever its size or security budget.
The sensible response is not alarm. It is a discipline: know which third-party applications are embedded in your software, understand what each one can reach, and plan for the day one of them has a bad day. That discipline is what a zero trust approach asks of security and IT leaders.
What happened recently, in short
On 16 November 2026, malicious JavaScript was served to visitors of around 178 organisations through Optimizely Web Experimentation, a tool used to test and personalise web pages. The figure comes from the security firm Report URI, which found it, and the incident has also been covered by The Register, BleepingComputer, The Record and TechTarget.
Optimizely's incident report says an unauthorised party exploited a flaw in the product's account sign-up flow to act as existing users, then edited the project JavaScript that customers' pages load. It says Optimizely detected the activity the same day, closed the flaw, removed the injected code from all affected projects and found no evidence that customer passwords were exposed. It adds that its other products were not affected and that sites with a strong Content Security Policy were protected. The company has engaged Mandiant to carry out an independent investigation.
As Report URI's summary puts it, an organisation can be compromised through a third-party dependency without its own code or infrastructure ever being breached. Some details are still to be confirmed: the report is preliminary, the forensic investigation is ongoing and the final root cause is yet to be published.
The point is not one supplier. Any vendor can have an incident, and the same lesson applies across the whole software estate. What matters is how much your organisation depends on third-party code without having looked closely at it.
Why third-party applications deserve a closer look
A typical enterprise web estate loads code from many outside companies: analytics, tag managers, personalisation and experimentation tools, chat widgets, consent platforms, payment providers and, increasingly, AI assistants. Each one either runs in your visitors' browsers or holds data about your customers and staff. Behind the front end, the same is true of SaaS platforms, plug-ins and APIs connected to your core systems.
The NCSC notes that, according to the 2025 Security Breaches Survey, very few UK businesses set minimum security standards for their suppliers. AI add-ons raise the stakes, because they often ask for broad permissions, such as reading mailboxes, documents or customer records, in order to be useful.
Zero trust in plain English
The NCSC describes zero trust as an architectural approach where inherent trust in the network is removed, the network is assumed hostile, and each request is verified based on an access policy. Applied to third-party software, that comes down to four habits:
- Do not trust by default. A supplier is not safe simply because it is well known or already inside your estate.
- Give the least access that works. Each application gets only the permissions and data it needs.
- Verify, do not assume. Check sign-in settings, permissions and change history, and test your assumptions.
- Plan for failure. Assume a supplier will one day be compromised, and limit the damage in advance.
A vendor audit checklist
This is generic good practice, not advice drawn from any single incident. Work through it for every third-party application embedded in your software, most business-critical first.
- Inventory. List every third-party application, plug-in, script, API and service, who owns it in your organisation, and why it is there. Include AI tools that teams have adopted on their own.
- Access and permissions. What can each supplier read, edit or send? Remove anything unused.
- Data held. What data does the supplier store, where and for how long? Is there a data processing agreement where the law requires one?
- Sign-in controls. Is multi-factor authentication enforced for every account that can change your site or configuration? Use named accounts rather than shared logins, remove leavers promptly and use single sign-on where it is offered.
- Contract and breach notification. How quickly must the supplier tell you about an incident, and who is your contact? What happens to your data if you leave?
- Logs and change history. Can you see who changed what, and when? Do you keep your own web and server logs for investigations?
- Code on your pages. Which scripts load in your visitors' browsers, and who can edit them? A Content Security Policy tells browsers which sources a page may load from. Optimizely's report also notes that cards entered in hosted payment frames were out of the script's reach, a useful reminder of why payment isolation matters.
A simple damage assessment
You do not need special tools. Take your most important supplier and answer six questions on one page.
- What could it reach? Pages, forms, customer records, payments, email, admin areas, internal systems.
- What could it do with that? Read, change, send or delete.
- How would we notice? Alerts, monitoring, a customer telling us, or nothing at all.
- How fast could we switch it off? Who holds the login, and can the supplier be removed from the site in minutes rather than days?
- Who would we have to tell? Customers, your card processor, regulators and the ICO, which expects notifiable personal data breaches to be reported where feasible within 72 hours of you becoming aware of them.
- What is the fallback? How the business carries on without it for a day.
Score each supplier high, medium or low for impact, then fix the high ones first. A font library that fails is an inconvenience. A changed script on your checkout page is a different matter.
Where to start
Three steps fit into a week. Write the inventory. Enforce multi-factor authentication wherever it is missing. Run the damage assessment for your top three suppliers. Then repeat the exercise twice a year, and whenever a team adds a new AI tool or application. Report the results to the board in plain terms: what is embedded, what it can reach and who owns the risk.
Join the conversation
How well do you know the third-party applications embedded in your software, and have you tested what would happen if one of them failed? We would like to hear what is working and what is not. Join fellow security and technology leaders from our CISO and CIO communities at an Inspired summit or roundtable to compare notes with your peers. Our events calendar shows where we are meeting next.
Sources
- Optimizely, "Security Incident Report: Web Experimentation", 17 November 2026: https://pages.optimizely.com/rs/947-RKE-048/images/Security%20Incident%20Report%20-%20Optimizely%20Web%20Experimentation%20-%20Nov%2017%202026.pdf
- Report URI, "World's Largest Brands Infected With Malware Via Optimizely", 16 November 2026: https://blog.report-uri.com/worlds-largest-brands-infected-with-malware-via-optimizely/
- The Register, "Optimizely hack turns world's biggest websites into malware dispensers", 17 November 2026: https://www.theregister.com/cyber-crime/2026/11/17/optimizely-hack-turns-worlds-biggest-websites-into-malware-dispensers/5310448
- BleepingComputer, "Optimizely supply chain attack infects major brands' sites with malware": https://www.bleepingcomputer.com/news/security/optimizely-supply-chain-attack-infects-major-brands-sites-with-malware/
- The Record, "Website management platform Optimizely confirms hackers put malicious code into customer sites": https://therecord.media/optimizely-malicious-code-customer-sites
- TechTarget, "Optimizely supply chain attack turns trusted scripts against 178 orgs": https://www.techtarget.com/news/optimizely-supply-chain-attack-turns-trusted-scripts-against-178-orgs/874997/
- NCSC, "Zero trust architecture design principles": https://www.ncsc.gov.uk/collection/zero-trust-architecture
- NCSC, "Supply chain security guidance": https://www.ncsc.gov.uk/collection/supply-chain-security
- ICO, "Personal data breaches: a guide": https://ico.org.uk/for-organisations/report-a-breach/personal-data-breach/
Image: padlock on a keyboard, via Unsplash (Unsplash License).


-p-500.jpg)

