Regulators Fine Failures in Routine Security Controls

Regulators Now Fine the Unglamorous Security Controls

Regulators do not need a clever attacker to find a violation. They need a gap between the controls an organization describes and the controls that existed on the day the data left.

The Italian data protection authority, the Garante, just showed what that gap looks like. In a decision dated 14 May 2026, which the European Data Protection Board posted on 10 November 2026, the Garante fined telecom operator Wind Tre S.p.A. EUR 1,715,600 for breaching GDPR Article 5(1)(f) and Article 32(1)(b). Two breaches, notified in February 2025, began at two retail outlets where attackers posed as technical support and talked employees into granting remote access to their devices. More than 365,000 customers were affected. For 41,359 of them, payment data was exposed as well.

Kiteworks was not involved in this incident, and nothing here claims otherwise. The social engineering call that opened the door is an identity and endpoint problem, and Kiteworks does not solve it. What the Garante fined was the second half of the story, the weak handling of certificates, private keys, credentials and API access that let a foothold in a retail outlet turn into mass extraction. That second half is the territory of Kiteworks secure data exchange and the Kiteworks Control Plane, and it is the part a CISO, a Chief Compliance Officer and a Data Protection Officer must be able to evidence on demand.

Key Takeaways

  1. The fine landed on the plumbing. The Garante penalized weak management of certificates, private keys and credentials plus loose API controls, not the phone call that fooled two employees.
  2. Security of processing is a list of verifiable behaviors. Article 32 asks for confidentiality, integrity, availability and resilience that you can show, which means evidence for every access.
  3. Social engineering is an identity problem, and data controls limit what follows. No data exchange platform stops an employee from being deceived, but bounded access decides how far an attacker can travel afterward.
  4. Custody beats inventory. Knowing where keys and certificates live is not enough, because regulators ordered secure storage, access controls and lifecycle processes.
  5. An incomplete audit trail is a second finding waiting to happen. If you cannot count who accessed what, your breach notification and your defense both rest on estimates.

The Garante Found Failures After the Foothold

The facts are worth stating carefully, because the headline fine hides the useful detail. According to the EDPB’s posting, the decision rests on two provisions, Article 5(1)(f), the integrity and confidentiality principle, and Article 32(1)(b), which concerns the ability to ensure ongoing confidentiality, integrity, availability and resilience of processing systems and services. The exposed payment data covered IBANs, postal payment slips and partially masked credit card numbers with expiry dates.

The technical findings read like an audit checklist. The Garante identified inadequate management and protection of digital certificates, private keys and access credentials in the company’s sales network. It found weaknesses in the security controls on certain application programming interfaces, which let attackers send a large number of requests and pull customer data in volume. And it found that Wind Tre’s own security assessments had missed vulnerabilities that deeper testing would have caught, according to the Garante’s newsletter announcing the decision on 16 July 2026.

The order is just as specific. Wind Tre must keep certificates in secure storage with access controls and safeguards against unauthorized export, put password management in place at its outlets, and formalize lifecycle processes for certificates and credentials. It must report the measures it took to the authority within 30 days of being notified of the decision. The Garante also credited prompt breach notification, corrective action after the incidents and cooperation with the investigation, which is a useful reminder that how you respond changes the number.

One housekeeping point. The EDPB carries this item as national news, a summary of the Italian authority’s decision rather than an EDPB ruling, so anyone building a legal position should read the decision itself.

A Complete Checklist of GDPR Compliance

Read Now

Security of Processing Is a List of Verbs

Most compliance programs treat Article 32 as a paragraph to cite. Read it as a list of verbs. The controller and the processor must implement technical and organisational measures appropriate to the risk, and the text names encryption, the ability to keep systems confidential, intact, available and resilient, and a process for regularly testing and evaluating whether the measures work. Article 32 never says “have a policy.” Every clause asks for something that happens and can be checked.

That is why enforcement keeps landing on basics. The CMS law firm’s GDPR Enforcement Tracker Report 2026, covering 2,685 analysed fines worth around EUR 6.11 billion with a data cut-off of 1 March 2026, ranks insufficient technical and organisational measures for information security as the third most common violation type. It sits behind legal basis and general processing principles and ahead of data subject rights.

Notice what that ranking implies. Security failures are not exotic. They are routine, and they are fined routinely, because the elements a regulator checks are ordinary ones. Who had access. Under what conditions. With which credentials. With what record. I call the result the plumbing penalty. Nobody gets a conference keynote for rotating a private key, and nobody avoids a fine for skipping it either.

So why do the basics stay unfinished? PwC’s 2027 Global Digital Trust Insights, a survey of 3,934 business and technology leaders in 71 countries fielded from May to July 2026, found that, among respondents asked about data foundations, organizations had implemented an average of only three of seven data risk measures across the organization, and just 5 percent had implemented all seven, down from 7 percent a year earlier. Those are leaders describing their own programs, which makes the gap harder to dismiss. Seven measures, three finished. That is a backlog, and a regulator reads a backlog as a finding.

Social Engineering Is Not a Data Control Problem

Here is where honesty matters. The way into Wind Tre was a deception. Attackers pretended to be technical support, and employees at two retail outlets let them onto their devices. Kiteworks does not stop that call from succeeding. Verifying who is on the phone, locking down remote support tooling, training retail staff, and vetting the identity of anyone requesting access all live in the identity and endpoint layers, and a data exchange platform is not the answer to them.

There are two lazy responses to a story like this, and both fail. The first is to blame the employees, schedule another awareness module and call the risk handled. Palo Alto Networks’ Unit 42 reported in its Global Incident Response Report 2026, built on more than 750 major incidents in over 50 countries, that identity weaknesses played a material role in nearly 90 percent of its investigations. People will always be the softest part of any identity layer, which means a program that depends on perfect employee judgment has already lost.

The second lazy response is the shopping trip, buying another detection tool because the last one missed something. The Garante did not fine Wind Tre for failing to detect a phone call. It fined the company for what an intruder could reach once inside. The variable that decides the size of the consequence is how far a single compromised session can travel, and whether anyone can prove what it touched. Assume the foothold will happen. Then design for the second move.

For a CISO, that reframes the question from “how do we prevent every deception” to “what is the maximum a deceived user can expose.” For a Chief Compliance Officer, it reframes the evidence request from “show us your training records” to “show us the access and the log.” That is a data-layer question, and it is where Kiteworks Control Plane capabilities apply.

Least Privilege Means Nobody Starts With Access

The first control regulators look for is almost embarrassingly simple. Does a user hold only the access the job requires? The decision does not describe Wind Tre’s internal permission model, so it would be unfair to guess at it. But it does describe an outcome, the extraction of customer data on a scale of more than 365,000 people through compromised retail access, and that outcome is what least privilege exists to prevent.

Kiteworks is built so that access controls start from zero. All users initially have no privileges. They receive functional privileges only when assigned a user profile, and they receive data access only by invitation from the people who manage that data. Administrators must explicitly enable elevated permissions. That default matters to an auditor because it flips the burden of proof. Instead of explaining why someone could reach a dataset, the organization shows who invited whom and when.

Separation of duties follows the same logic. The platform ships with eight default administrator roles, plus custom roles, so duties can be split between, for example, helpdesk staff who manage accounts and investigators who run eDiscovery. A DPO answering a supervisory authority wants that kind of boundary documented, since it shows that administrative power and data power are not automatically the same thing.

Least privilege will not stop a deceived employee from using the access they legitimately hold. What it does is cap the damage at that employee’s own scope, and the difference between a scoped incident and a 365,000-person incident is the difference between a notification and a fine.

Attribute-Based Access Beats Role Lists at Scale

Roles answer a blunt question. Is this person a retail agent, an analyst, an administrator? That works until the role is stolen. The better question is whether this person, in this situation, should touch this particular data in this particular way. That is what ABAC answers, and it is why mature programs layer it over RBAC instead of choosing one.

In Kiteworks, the combination of RBAC and ABAC is called the data policy engine. ABAC policies, called data policies in the product, govern each data access dynamically based on attributes of the data, the user and the attempted action, evaluating conditions such as folder path or tag, user domain and what the person is trying to do. Each rule applies a directive chosen by the policy author. The same engine applies across email, file sharing, managed file transfer, forms and APIs, so a policy is written once rather than reinvented per channel.

There is a location control as well. Sign-ins can be restricted by geofencing at the individual user, user profile or system level, and administrators can set block lists and allow lists for IP address ranges. A compromised credential used from an unexpected place meets a policy before it meets a file.

None of this is magic, and I want to be careful about scope. These controls govern access to content that lives in or moves through Kiteworks. They do not rewrite the permission model of a telecom’s customer relationship system. What they give a compliance leader is a precise, reviewable policy statement for the data in their care, which is the thing Article 32 asks them to be able to produce.

Key and Credential Custody Is Where Fines Hide

The Garante’s order on certificates and keys deserves more attention than it got. Inadequate management and protection of digital certificates, private keys and access credentials is not a headline-friendly finding, yet it is the sort of failure that turns a single stolen secret into broad access. The corrective measures, secure storage with access controls, safeguards against unauthorized export and a lifecycle process, describe custody, not inventory.

Custody is the word to hold onto. Knowing that a private key exists is inventory. Knowing who can read it, export it, rotate it and retire it, and being able to show the record, is custody. A regulator will accept a gap you have documented and are closing far more readily than a gap you never knew you had.

Within its own scope, Kiteworks supports custody in several documented ways. Customers hold their own customer-controlled encryption keys, so Kiteworks staff, and outside actors such as governments in the case of blind subpoenas, cannot decrypt a customer’s private data. The platform integrates with hardware security modules, including SafeNet Luna Network HSM from Thales and AWS Key Management Service, with support for Entrust HSMs, which keeps key management outside the appliances. Strong sign-in includes MFA options such as RADIUS, PIV and CAC cards, time-based authenticators and one-time passwords, plus certificate-based authentication that notifies administrators before a root or intermediate certificate expires.

Now the limit, stated plainly. These capabilities protect the keys and sign-in credentials for content in Kiteworks. They do not inventory, rotate or vault every certificate across a retail network, a billing platform or an internal API estate. Wind Tre’s problem lived across a sales network, and no single product closes that for an enterprise. The honest use of Kiteworks here is to make sure that the sensitive content you exchange never depends on a secret that nobody governs.

API Access Needs the Same Policy as a Person

The most expensive sentence in the EDPB summary may be the one about APIs. Weak controls on certain APIs allowed a large number of requests, and customer data came out in bulk. APIs are where data protection programs most often discover that machine access was never governed the way human access was, because most access reviews start with people and skip integrations.

The principle for a zero trust data protection program is simple. Every caller, human or programmatic, proves who it is, is checked against policy for the specific request, and leaves a record. Kiteworks exposes REST APIs for custom integrations with authentication options that include OAuth, MFA and certificate-based authentication, and those integrations fall under the same policy engine and the same audit log as everything else. An integration is not a side door around governance.

Rate limiting deserves a precise statement because it is easy to overstate. Kiteworks product documentation covers configurable rate limiting, with request throttling per user per time period, for the Kiteworks Secure MCP Server, the component that connects AI assistants to Kiteworks content. It does not describe the same feature for every API, so this post does not claim it. If bulk extraction through a general API is your concern, ask your vendor, and ask us, exactly which limits apply to which interface.

What CISOs can do today is the unglamorous part. List every API client that can reach personal data. Record its owner, its credential, its scope and its call volume. Anything without an owner is the next paragraph in someone else’s decision letter.

An Audit Trail Is Evidence Only When It Is Complete

Look at the precision of the numbers in this case. More than 365,000 customers affected. Exactly 41,359 with payment data exposed. Those are not estimates, and the second figure matters because payment data raises the stakes for the customers involved. Producing a number like that depends on knowing who accessed what, and that depends on logs that kept everything.

Here is where many organizations are quietly exposed. Kiteworks maintains a single consolidated audit log that records user authentication, successful and failed, file views, downloads, uploads, edits, deletions, sharing, permission changes and configuration changes across channels. Kiteworks does not throttle logging, so entries are captured in full during surges of activity, and log feeds reach a SIEM through syslog or the Splunk Universal Forwarder. The built-in CISO Dashboard visualizes anomalies such as unusual download locations and traffic levels.

The value to a compliance leader is the chain of custody it produces. Every file, every access, every action carries who, what, when, where and how, so the history of any sensitive file can be rebuilt for a regulator, a legal hold or an investigation. That is what turns a log from a technical artifact into evidence.

Does the audit burden bite in practice? Thoropass’s 2026 State of Audit and Compliance Report, a vendor survey of 536 security and compliance leaders, reported that 53 percent struggle to collect evidence across multiple tools and that 38 percent are not fully confident they would pass an unplanned audit. Treat it as directional, since Thoropass sells audit software and the report does not disclose fielding dates or a margin of error. Still, the pattern matches what the Garante implied. When evidence lives in five places, the regulator’s clock runs faster than the team’s search.

Where Kiteworks Fits and Where It Does Not

A table helps keep the boundary honest. The left column is what the Garante found or ordered. The right columns say what Kiteworks supports and what remains outside it.

What the Garante found or ordered What Kiteworks supports What stays outside
Social engineering gave attackers remote access to employee devices MFA and certificate-based authentication for access to Kiteworks content Help desk verification, remote support tooling and staff training
Inadequate management of certificates, private keys and credentials Customer-owned keys, HSM integration and certificate expiry notices for sign-in Certificate and key inventory across non-Kiteworks systems
Weak API controls allowed mass extraction Policy-governed REST APIs and rate limiting on the Secure MCP Server Controls inside other organizations’ own APIs
Order for password management and secure storage Least-privilege defaults, RBAC and ABAC data policies Password policy for systems Kiteworks does not touch
Security assessments missed vulnerabilities A complete audit log, SIEM feeds and analytics for anomalies The vulnerability testing program itself

Read the right-hand column as carefully as the middle one. A vendor that tells you it solves all five rows is selling a story, and the Chief Compliance Officer who repeats it to a regulator owns the consequences. For a DPO preparing a response, the language to use is that these controls support the demonstration of appropriate security under Article 32. No product makes an organization compliant. Kiteworks helps demonstrate controls, and people still run the program.

What to Do Monday Morning

Start with the question the Garante’s numbers invite. Pick your largest customer dataset and ask who can export it in bulk today, how they would authenticate, and where the record lives. If the answer takes more than a day to assemble, you have found your first finding, so write it down before a regulator does.

Next, treat keys and certificates as custody problems with owners. For every certificate and private key that protects personal data, name an owner, a storage location with access controls, a rotation date and an export restriction. This is the work the Garante ordered Wind Tre to formalize. It costs far less to do on your own schedule than on a 30 day deadline.

Third, inventory machine access. List every API client and integration that can read personal data, then confirm each one is authenticated, scoped to the minimum, governed by the same policy as a human user and visible in the same log. Anything that fails goes onto the remediation list with a date.

Fourth, rehearse the evidence request. Choose a cohort of customers, ask your team to show every access to their records over the last 90 days, and time the exercise. The GDPR compliance expectation is that you can answer in days, and your DPO should own the stopwatch.

Last, be honest about the boundary. Fix the social engineering entry point with the identity and endpoint owners, because it is a separate program with separate controls. Then make sure the data layer behind it is built so that a single deceived employee cannot expose a customer base.

To learn more about producing audit-ready evidence for GDPR security of processing, schedule a custom demo today.

Frequently Asked Questions

Article 32 requires controllers and processors to implement technical and organisational measures appropriate to the risk, and it names encryption, the ability to maintain confidentiality, integrity, availability and resilience, and regular testing of whether the measures work. The text does not prescribe specific tools, so the standard is whether your measures fit the risk and whether you can show they operate. Regulators translate that into evidence requests about access, credentials, keys and logs. A good starting point is mapping your GDPR compliance evidence to each clause, and using a data policy engine to document who can reach what.

A successful social engineering attack does not by itself determine a fine, but the controls around it can. In the Wind Tre case, the Garante fined the company for weak management of certificates, private keys and credentials and for inadequate API controls, not for the fact that employees were deceived. Kiteworks does not prevent social engineering, since that sits in the identity and endpoint layers, but least privilege and access controls limit how far a compromised user can reach. This is general information, so confirm your exposure with your DPO and legal counsel.

Expect a request to show who accessed the affected data, how they authenticated, what security measures were in place and what you did after discovery. The Wind Tre decision shows regulators weighing prompt notification, corrective action and cooperation alongside the technical failures. A consolidated audit log with complete entries and a documented chain of custody makes those answers fast. Gaps in logging tend to become a second problem, because they make the affected population hard to count.

You prove it with a log that captures every authenticated request and a policy engine that shows what was allowed. Kiteworks records authentication, file views, downloads, uploads, edits, sharing and permission changes in one audit log and feeds it to a SIEM through syslog or the Splunk Universal Forwarder. REST API integrations fall under the same policy engine and the same log as human users. If you also need volume limits, check which interface they apply to, since the documented rate limiting covers the Secure MCP Server specifically, and review the SIEM integration with your SecOps team.

Start by finding out who can export your largest customer dataset in bulk and whether you can show every access to it within a day. Then assign an owner, storage location, rotation date and export restriction to each certificate and key that protects personal data, and list every API client with access to personal records. Treat each gap as a dated remediation item, because documented progress reads better than undiscovered weakness. Your DPO should co-own the exercise, and customer-controlled encryption keys are a good place to begin the custody conversation.

Additional Resources

Get started.

It’s easy to start ensuring regulatory compliance and effectively managing risk with Kiteworks. Join the thousands of organizations who are confident in how they exchange private data between people, machines, and systems. Get started today.

Share
Tweet
Share
Explore Kiteworks