Explore our new Cybersecurity Terms Glossary, explained by our experts.

View the Glossary View the Glossary
PRIVIA

Cybersecurity Terms Glossary

How Do You Verify a Data Leak?

Someone claims to be selling your data. Before you notify anyone, you have to establish whether the sample is genuine, whether it came from you, and whether it is new.

Illustration for the cybersecurity term Data Leak Verification
Home / Blog / Cybersecurity Terms / How Do You Verify a Data Leak?
17 November 2026

How Do You Verify a Data Leak?

How Do You Verify a Data Leak?

Claims arrive in several ways: a post on a criminal forum, an email demanding payment, a journalist’s enquiry, or an automated alert from a monitoring service. All of them share the same problem. The claimant has an incentive to exaggerate, and your first obligation is to establish facts quickly, because a regulatory clock may already be running.

Start With the Sample, Not the Claim

Almost every genuine claim comes with proof: a few hundred rows, a directory listing, or a handful of documents. Work from that rather than from the description. Three questions follow in order.

Is the data real? Pick a dozen records at random and look them up in your own systems. If the names, addresses, and identifiers match live records exactly, including unusual values and known data quality quirks, the data is real. Fabricated samples typically fail on internal consistency: dates that do not align, identifier formats that look plausible but do not match your sequence, or postcodes that do not correspond to the addresses.

Is it yours? This is the question most often answered wrongly. Field structure is the strongest evidence: your table has a column other organisations would not have, your customer reference format is distinctive, or the sample includes an internal flag that only your application writes. Seeded canary records are better still, which is why organisations that have planted a handful of fictitious entries in each exported dataset can usually answer this in minutes rather than days.

Is it new? Recycled breaches are common. Check whether the records correspond to a period your systems no longer hold, whether the password hashes use an algorithm you stopped using years ago, and whether the dataset matches a known historical breach of a supplier. Have I Been Pwned and similar services can confirm whether a dataset has circulated before. The 2016 sale of Dropbox credentials turned out to be data taken in 2012, and the same pattern recurs constantly.

Look for the Route Out

If the data is yours, the next question is how it left, and the answer is frequently not a breach of your own systems. Candidates, roughly in order of frequency: a compromised third party processing data on your behalf, a misconfigured cloud storage bucket, an employee export to personal storage, a compromised mailbox containing attachments, a public API returning more than intended, and only then an intrusion into your core systems.

Narrowing it down depends on what the sample tells you. The fields present, the date range, the record count, and the ordering all indicate which system produced the export. A dataset that stops in March and includes exactly the fields in your reporting extract points somewhere quite specific. Preserve the sample, its source, and the time you received it, because the chain of custody matters if law enforcement or litigation follows.

The Clock Question

Under Article 33 of the UK GDPR, the 72-hour notification period to the Information Commissioner’s Office runs from the point at which you become aware of a personal data breach, and awareness means a reasonable degree of certainty that a breach has occurred, not completed investigation. A credible sample that matches your records usually establishes that. The ICO accepts phased notification: report what you know within the window and supplement it as the investigation proceeds. Delaying notification until the scope is fully understood is the mistake that turns an incident into an enforcement matter.

Dealing With the Claimant

Engage carefully and never alone. Agree in advance who may communicate with an extortionist, keep the technical team out of negotiation, and involve legal counsel and law enforcement early. UK organisations should report to Action Fraud and, for significant incidents, to the NCSC. Be aware that payment may engage sanctions legislation depending on who is behind the demand, which is a legal question rather than a security one.

Ask for additional proof covering a different part of the dataset. A claimant who can only ever produce the same 200 rows is usually working from a sample they bought rather than a full copy, and that materially changes your assessment.

At Privia Security, we handle leak verification as part of our digital forensic incident response services, including the source attribution work that decides which system actually leaked. Preparing in advance helps more than any tooling: seed canary records now, and keep a register of which exports contain which fields, so that the attribution question has an answer before you need it.

FAQ – Frequently Asked Questions

How Can You Tell Whether Leaked Data Came From Your Organisation?

Compare the field structure rather than the content. Column names, identifier formats, internal status flags, and the exact combination of fields present are far more distinctive than names and addresses, which many organisations hold. Seeded canary records are definitive: fictitious entries unique to a specific export immediately identify both the source system and, if seeded per recipient, which third party lost the data.

Does the 72-Hour Clock Start When the Claim Is Made?

It starts when you have a reasonable degree of certainty that a personal data breach has occurred, not when a claim first reaches you. A short period of verification before that point is accepted, but it must be short and diligent. Once a sample matches your live records, awareness is generally established, and the ICO expects a phased report rather than a delayed complete one.

What If the Data Turns Out to Be Old or Recycled?

Document the comparison that proves it: matching record counts against a previous incident, hash algorithms you no longer use, or a date range ending before a known migration. If the data relates to an earlier breach that was already notified, you may not need to notify again, but record the assessment. If it relates to a breach never identified, you have a new incident regardless of its age.

Should You Download the Sample From a Criminal Forum?

Only through a controlled process agreed with legal counsel, using an isolated system, with the download and its provenance logged. There are legitimate reasons to obtain the sample and real risks in doing so, including malware, further exposure of personal data, and questions about how your organisation came to possess it. Many organisations route this through an external provider for exactly those reasons.

Related Terms

You May Be Interested In These

What Are Reflection and Amplification Attacks?

Cybersecurity Terms

17 November 2026

What Are Reflection and Amplification Attacks?

Read More
What Is a Safety Instrumented System?

Cybersecurity Terms

17 November 2026

What Is a Safety Instrumented System?

Read More
What Is Telecom Fraud and Revenue Leakage?

Cybersecurity Terms

17 November 2026

What Is Telecom Fraud and Revenue Leakage?

Read More