NOV 20, 2026

Your Data Is Only as Safe as the Companies You Share It With

Third-party cybersecurity risks are the risks that customer, employee or business information is exposed because an outside provider that holds or processes it is attacked or misconfigured. The provider could be a cloud host, payment processor, marketing platform, software vendor or AI tool. Your own systems can stay intact while your data still leaks, which is why protecting information now means controlling what you share, not only what you secure.

Third Party Cybersecurity Risks Who Protects Your Data

Key takeaways

  • On 19 November 2026, JPMorgan Chase, Visa and Mastercard each said outside service providers had been targeted. Visa and Mastercard said payment data and payment networks were not affected, and the exposed information was reported to be marketing and benefit-programme data such as names and email addresses.
  • The European Commission confirmed an AI-generated cyber campaign detected on 22 October. It says it has no evidence of a staff-wide data breach, and its investigation is ongoing.
  • Exposure also comes from everyday habits: uploading contracts, customer records and chat transcripts to external tools, including AI assistants.
  • The most effective defences are simple: keep an inventory of providers, share only what each one needs, enforce least-privilege access and MFA, and bring vendors into your incident response plan.
  • Anonymising or redacting sensitive details before they reach AI applications reduces what a later breach or retention surprise can expose. It is an added privacy layer, not a replacement for security controls.

Picture an ordinary Tuesday at a mid-sized company. Sales exports a customer list to a marketing platform. HR keeps employee records in a cloud system. A payment provider handles card transactions. A finance analyst drags a spreadsheet of account details into an AI assistant to “just tidy it up,” and a lawyer pastes a draft contract into another tool to get a summary before lunch.

Nobody did anything reckless. Every one of those services is probably reputable, contractually approved and protected by a serious security team. And yet a fair question hangs over the day: where is all of that information now, who can reach it, and what happens if one of those companies is breached?

That question is the heart of third-party cybersecurity risks, and last week gave it a vivid illustration. On 19 November 2026, JPMorgan Chase, Visa and Mastercard each acknowledged that outside service providers had been caught up in cyber incidents. None of the three said its own core systems had been broken into. That detail is exactly why the story matters.

A security breach does not always start inside your company

Most organisations picture a breach as something that happens to them: an attacker finds a way past the firewall, into the network, onto the servers. Third-party exposure works differently. The attacker never touches your systems. They go after a company that holds a copy of your data, or a slice of it, and your information leaves through the side door.

That copy exists because modern business runs on outsourcing. Cloud infrastructure hosts the applications. Payment processors handle transactions. Marketing and customer management platforms hold contact lists and campaign data. Software vendors ship code that runs inside your environment, and a long tail of data processors and technology partners handle everything from payroll to document signing.

Each relationship is sensible on its own. Together they form a map of dependencies that few organisations can draw from memory. JPMorgan’s own annual reports have described the problem plainly for years: vendors, supply chain partners and other third parties are themselves a source of cybersecurity risk, and incidents at those parties could result in lost or compromised information belonging to the firm or its clients. The risk is not new. What is new is how many places a single company’s data now lives.

The practical consequence is uncomfortable. A business can patch diligently, train its staff and pass its audits, and still see customer information appear in a leak because a supplier it barely thinks about was targeted. This is why third-party data breaches and vendor cybersecurity risks belong in the same conversation as endpoint protection and password policy rather than in a procurement footnote.

What the November incidents reveal

Start with what has been confirmed, because the details matter and several early headlines ran ahead of them.

The US financial sector, 19 November. According to ABC News, JPMorgan said that third parties providing services to the bank were the target of a cyberattack. The bank did not say when it happened or name the companies involved. Per reporting from the Wall Street Journal as summarised by PYMNTS, JPMorgan believes attackers accessed marketing data that could include customer names and email addresses.

Visa said the incident involved a third-party service provider in the Asia-Pacific region, and a spokesperson told the Journal that payments data such as card numbers was not affected, nor were Visa’s own systems. Mastercard said a third-party provider had suffered an incident that exposed a limited amount of non-payments information, specifically email addresses from a small subset of cardholders enrolled in a particular benefit programme. Mastercard added that the exposure had been contained and that its payment network and cardholders’ ability to use their cards were unaffected.

Three things were not established. None of the companies named the affected providers. Public reporting does not confirm whether the three incidents involved the same vendor. And the claim that a ransomware group called n0n was behind the leak comes from the group itself and from security outlets such as Cybernews, not from the companies, which did not confirm the attacker’s identity. Reports that data on thousands of customers was leaked also come from media coverage rather than from company disclosures. JPMorgan separately attributed a recent service outage to a software bug and has said it found no connection between that outage and the third-party incident.

So this was not a story about payment rails failing. It was a story about marketing lists and benefit-programme data, held by suppliers, ending up in the wrong hands. For the people on those lists, the distinction between “payment data” and “email address” is real but not entirely comforting: names and email addresses are precisely what fuels convincing phishing messages that appear to come from the bank or card network.

The European Commission, detected 22 October. The second incident is different in kind, and it is worth being precise about why. Speaking on 19 November, Commission spokesperson Thomas Regnier confirmed an “AI-generated cybercampaign” detected on 22 October. As a precaution, the Commission temporarily took several applications offline, and most have since been restored. The investigation is ongoing. Regnier said there was no evidence of a data breach affecting Commission staff as a whole, declined to say who was behind the campaign, and said he was not aware of any ransom demand.

That is where the confirmed record ends. The Commission has not publicly detailed how AI was used or which systems were affected beyond the applications it took down. Politico reported that systems for payslips, time management and medical reimbursements were offline during a roughly ten-day disruption and that large amounts of personal data may have been compromised. That is a press report of a possibility, and it sits alongside the Commission’s own statement that it has no evidence of a staff-wide breach. Both can be true while the inquiry continues, and neither should be presented as settled.

This is not a vendor incident, and it would be wrong to dress it up as one. Its relevance is the tempo it implies. If attackers can use AI to scale and speed up their work, the window between a weakness appearing somewhere in your supply chain and someone exploiting it gets shorter. Your vendors’ patching discipline then becomes part of your own exposure.

Taken together, the two stories say something uncomfortable about how data security actually works now. Organisations are held accountable by customers for information they no longer fully control. Their disclosures often come in cautious fragments, with the supplier unnamed and the timeline vague. And the fastest-growing category of data leaving the building is no longer only files sitting in a vendor’s database. It is also what people type into the AI tools in front of them.

The information you share can become your biggest risk

The incidents above involved data that organisations deliberately handed to suppliers. A lot of exposure happens more casually, through habits that never appear in a vendor register.

Consider four situations that will be familiar to most teams:

  • A customer support agent pastes a long chat transcript into an AI assistant to draft a reply. The transcript includes the customer’s full name, home address and part of an account number.
  • A finance team uploads a batch of invoices and bank details to an external processing platform that someone found online and signed up for with a work email.
  • A legal team asks an AI tool to summarise a confidential contract, complete with party names, pricing and termination terms.
  • An employee forwards a customer export to an outside agency to help with a campaign, without checking how long the agency keeps the file or who else can open it.

None of these people is behaving maliciously. Each is trying to finish a task faster. But in every case, information has moved beyond the organisation’s direct control, and the questions that matter afterwards are ones nobody asked at the time. How long is it stored? Who at the provider can see it? Is it used to improve a model? What happens to it if the provider is breached, acquired or simply changes its terms?

This is not an argument that outside services are unsafe. Many are better secured than the average in-house system. The argument is narrower: trust is not a control. A provider you trust still needs to be visible to you, limited in what it receives and accountable in writing for what it does with it. Without those three, trust is just an assumption waiting to be tested.

Why AI adds another layer to the data privacy problem

Generative AI has changed how information leaves organisations, mostly by making it frictionless. Sharing a document with a vendor once required an email, a contract and often someone’s sign-off. Pasting the same document into a chat window takes two seconds and feels like using a calculator.

Several things follow from that. Prompts and uploaded files routinely contain material that would never have been sent to an outside party in the past: customer records, internal strategy notes, source code, HR cases. Retention and processing practices vary widely between tools and even between plans of the same tool, and employees rarely read the fine print. Many AI applications also sit on third-party infrastructure or connect to other services, so a single prompt can touch several companies you have never assessed.

Then there is the problem of reversal. Once information enters an external AI workflow, there is usually no reliable way to pull it back. You cannot recall a prompt the way you might ask a vendor to delete a file, and you may not know which logs, caches or downstream integrations hold a copy. This is what separates AI data privacy from older vendor risk: the sharing is more frequent, more informal and harder to undo.

The sensible response is not to ban AI, which tends to push use underground. It is to decide what kinds of information may go into which tools, and to reduce what is sent in the first place.

How businesses can reduce third-party data exposure

Effective third-party risk management is less about a perfect questionnaire and more about a handful of habits applied consistently. Here is what tends to matter most, and why.

  1. Keep a living inventory of providers and the data each one touches. You cannot assess what you cannot see. Even a simple register of vendors, the categories of information they hold and the business owner responsible changes the quality of every later decision. Include the tools employees adopted on their own.
  2. Share only what a provider genuinely needs. Data that was never sent cannot be leaked. If a marketing platform needs email addresses and first names, it does not need dates of birth or account numbers. This is the cheapest control available and the most often skipped.
  3. Read the security, privacy and retention terms. Look for how long data is kept, whether it is used for training or analytics, which subprocessors have access, and how quickly the vendor commits to notifying you of an incident. Slow notification turns a contained problem into a public one.
  4. Apply least-privilege access. Give people and integrations the minimum permissions required, and remove them when the project ends. Many third-party incidents become serious because a connection had far more reach than its purpose required.
  5. Monitor transfers and review access regularly. Quarterly reviews of who and what can reach your data catch forgotten integrations, former contractors and stale API keys before an attacker does.
  6. Require multi-factor authentication and strong authentication. Stolen credentials remain one of the most common ways into a vendor’s systems and then into yours. Require MFA from vendors where it is possible to do so, and enforce it on your own side.
  7. Keep software and integrations updated. If attackers are using AI to find and exploit known weaknesses faster, delay is a larger risk than it used to be. Ask vendors how quickly they patch critical vulnerabilities.
  8. Write clear rules for AI tools and confidential information. People need practical guidance, not a vague policy: which tools are approved, which categories of data are off limits, and what to do when unsure.
  9. Anonymise or redact sensitive details before sharing with external AI applications whenever appropriate. Removing identifiers at the source means a later breach or retention surprise exposes far less.
  10. Include third parties in your incident response plan. Know who to call at each vendor, what they are obliged to tell you, and how quickly you must inform your own customers. The JPMorgan, Visa and Mastercard statements show how often an organisation’s first move is to learn the details from a supplier.

None of this requires a large budget. It requires someone owning the list and a habit of asking the question again each time a new tool arrives.

Protect sensitive information before it reaches AI applications

Most conversations about AI risk focus on which tools employees use. That matters, but it leaves out a second question: what do those tools actually receive? If the answer is raw customer conversations, contracts and records, then the safety of the whole workflow depends on every provider in the chain behaving perfectly forever.

This is where anonymisation and redaction earn their place. Instead of trusting an AI service with identifiable information, you remove or replace the identifiers first, so the service works with a version of the content that cannot easily be tied back to a real person or company.

Questa AI is built around this idea. It anonymises or redacts personal and confidential information in documents and prompts before they are passed to an AI model, and offers cloud and on-premise deployment options. It is designed as an additional privacy layer for AI workflows, not a substitute for your firewalls, vendor due diligence or incident response, and it does not protect systems that sit outside the workflow it covers.

A short illustrative example shows the principle. Imagine a support team wants an AI assistant to summarise this message:

“Hi, I’m Emily peter. My card ending 4821 was charged twice at Alder & Finch Ltd on 14 November. Please call me on 07700 900123 or email priya.nair@example.com.”

After anonymisation, the assistant might receive something closer to this:

“Hi, I’m [PERSON_1]. My card ending [CARD_1] was charged twice at [MERCHANT_1] on [DATE_1]. Please call me on [PHONE_1] or email [EMAIL_1].”

The assistant can still summarise the complaint, identify the double charge and suggest a reply. What it never sees is who the customer is. If that provider were later breached, or kept the prompt longer than expected, the exposed text would be of little use to an attacker.

Before sharing sensitive business information with AI, consider what should be protected first. Explore how Questa AI can help your organisation introduce data anonymisation into everyday AI workflows.

Data protection must extend beyond your own systems

The uncomfortable lesson of November’s incidents is that a company can do almost everything right inside its walls and still read about its customers’ details in a leak. No organisation can audit every supplier’s infrastructure or predict which one will be targeted next.

What it can control is the decision that comes before any of that: what to hand over, to whom, in what form, and with what rights to ask questions afterwards. Companies that make those decisions deliberately will still face incidents, but the incidents will expose less.

You will never control every system your data passes through. You can still decide what leaves with it.

Frequently asked questions

They are the risks that arise when an outside provider that holds, processes or has access to your data is attacked or misconfigured. Examples include cloud hosts, payment processors, marketing platforms, software vendors and AI tools. Your own systems may remain intact while your information is still exposed.

Yes. If a supplier stores a copy of your customer or employee data, a breach at the supplier can expose it. In the November 2026 financial-sector incidents, the companies said the affected parties were their service providers, and Visa and Mastercard said their payment data and networks were not affected.

Start with a vendor inventory covering procurement records, finance payments, single sign-on and application logs. Then ask each team which tools they use to process customer or employee data, including free or personal-account tools adopted without approval. Record what data each provider holds and who owns the relationship.

It depends on the tool, its contract, its retention and training terms, and what you put in. The safest default is to share the minimum necessary and to anonymise or redact personal and confidential details first, so the tool can do its job without receiving identifiable information.

Map your vendors and the data they hold, cut down what you share to what each one actually needs, and enforce MFA and least-privilege access. Then make sure your incident response plan names a contact at each critical provider and sets expectations for notification.

Conclusion

By the companies’ own accounts, November’s financial-sector incidents did not touch payment systems. The exposure came from data held one step removed, by suppliers that customers never chose and rarely heard of. Third-party risk will not be solved by one tool or one policy, because it lives in dozens of small decisions about what gets shared and with whom.

A realistic first step this week is to list the five providers that hold your most sensitive information, check what each one actually receives, and set a plain rule for what may be pasted into AI tools. Where identifiable details are not needed for the task, anonymise or redact them first. You will still be affected when someone else is breached, but you will be able to say what was exposed, and what never left.

Abhi Author

About the author:

Abhiroop Sharma

Ex. Distinguished technology leader

Distinguished technology leader with 18+ years of progressive experience spanning AI, Web3, SaaS, eCommerce, and blockchain governance. Demonstrated success in driving digital transformation across global markets, with expertise in scaling enterprise solutions from concept to implementation. Proven track record of reducing implementation timelines by 50% and building high-performing teams across multiple organizations. Currently focused on pioneering AI implementation and Web3 integration strategies for emerging technology ventures.
Follow the expert:

Related Articles

View More
AI Risk Management: How Enterprises Control AI Risk
SEP 23, 2026
Privacy Cafe

AI Risk Management: How Enterprises Control AI Risk

AI risk management means knowing where AI touches sensitive data, judging what could go wrong, and putting real controls in place before it does.

Read More
AI in Cybersecurity: Risks, Defenses and the IMF Warning
MAY 08, 2026
Privacy Cafe

AI in Cybersecurity: Risks, Defenses and the IMF Warning

AI strengthens both attackers and defenders in cybersecurity. The IMF's 2026 warning shows why speed and infrastructure now decide enterprise cyber risk.

Read More
AI Treasury Risk: Cyber Threats, Assessment & Controls
MAY 04, 2026
Privacy Cafe

AI Treasury Risk: Cyber Threats, Assessment & Controls

Treasury's AI cyber warnings, the FS AI RMF's 230 control objectives, and a practical risk assessment framework for banks and treasury teams.

Read More