Almost every confidentiality clause is written the same way, and it is squeezed from both sides. Define the protected material too widely and a court may decline to enforce it. Define it too narrowly and the thing that mattered was never covered. The drafter is guessing, before anyone knows what will be shared.
There is a way out that does not involve better adjectives. Define the scope by reference to the documents themselves, using their cryptographic fingerprints, so the agreement points at identified files rather than describing them.
| Approach | The difficulty | Where it fails |
|---|---|---|
| Everything disclosed is confidential | May be treated as unreasonable or impractical to comply with | Hard to enforce, and gives no help identifying what a claim actually relies on |
| Only material marked confidential | Depends entirely on marking discipline under time pressure | Anything shared unmarked falls outside the agreement |
| A written list of categories | Written before anyone knows what will be shared | Material that mattered was never in a listed category |
| A schedule of certificate references | Requires depositing each document before it is shared | Covers only what was recorded, so the discipline has to hold |
The last row has a cost of its own, stated plainly: it only covers what was recorded. If a document is shared without being deposited, the schedule does not reach it. That is a process discipline, not a technology, and it is worth knowing before adopting the approach.
The requirement is particularity. A party asserting confidential information has to identify it with enough specificity that a court can tell what is protected and frame a remedy around it. That requirement does not sit in one numbered provision. In the EU it follows from the definition of a trade secret in Article 2(1) of Directive (EU) 2016/943 read with national rules of procedure, the same applies in the UK under the regulations implementing the Directive, and under the US Defend Trade Secrets Act a claimant must identify the secret with reasonable particularity.
English proceedings show what that means in practice. In IlliquidX Ltd v Altana Wealth Ltd and others [2024] EWHC 2191 (Ch), an application to amend particulars of claim, the court restated that claims for breach of confidence and misuse of trade secrets must be properly particularised both as to the information relied on and as to the misuse alleged, and refused some of the proposed amendments. IlliquidX went on to win at trial and to hold that win on appeal, so this is not a story about a claim failing. Our post on what standard disclosure cannot prove follows the case through to the costs judgment, where the lack of precision still cost the winning party.
The point holds whichever way a case goes. The material has to be identifiable before the dispute, in terms specific enough to survive being read back by an opponent.
A confidentiality clause describes the protected material because the material cannot go in the agreement itself. That is the whole difficulty: the description has to be broad enough to cover what will be shared and precise enough to be enforced, and those pull in opposite directions.
A fingerprint is neither broad nor narrow. Each document is deposited before it is shared, which produces a certificate carrying the file name, the SHA-256 fingerprint, a timestamp from an accredited trust service provider and a blockchain anchor. The certificate reference goes into a schedule to the agreement. The obligation then attaches to identified documents, and the schedule reveals nothing about their contents, because a hash cannot be reversed.
When documents change during a negotiation, each version gets its own deposit and reference. Whether that is worth doing, and how the clause should read, is a question for the lawyer drafting the agreement. This is a drafting approach rather than a legal rule, and nothing here is advice on your agreement.
Defining what an agreement covers and proving what was actually sent are two different problems, and certificate services address only the first. In a dispute the second is usually where the argument lands: not whether the category was covered, but whether this document reached that person on that date.
A sent folder is held by the party relying on it. A recipient download log is held by the opposing party. Neither is independent, and a court will treat both accordingly. Provlyn shares from a vault instead: each recipient opens the material through a controlled link, each view is logged against a timestamp, and access can be withdrawn. Each recipient can be issued an individually watermarked copy, so a document that surfaces where it should not identifies who received it.
Where the material also qualifies as a trade secret, those access records do double duty, since keeping information access-restricted is one of the reasonable steps a holder has to evidence.
A page about making confidentiality agreements work should say where they stop. In the UK an NDA cannot be enforced against someone who is, or reasonably believes they are, a victim of crime, to prevent disclosure to the police or other bodies investigating or prosecuting crime, to a qualified lawyer for advice about it, or to a regulated professional for support. An agreement that purports to do so is unenforceable to that extent.
Confidentiality provisions also cannot be used to conceal unlawful behaviour, and the law in this area has been moving. None of it affects the ordinary commercial use of an NDA between businesses, which is what this page is about. It matters because a confidentiality regime built on the belief that an NDA silences everything is built on a false premise, and no amount of evidential infrastructure fixes that.
One automated sequence, each step taking the output of the one before it. The document is hashed with SHA-256. That hash is timestamped under RFC 3161 by an accredited trust service provider. The timestamped hash is anchored to the Bitcoin blockchain through OpenTimestamps. Only the hash travels, so the document stays where it is and the schedule can be shared with the other side without disclosing anything.
eIDAS qualification is a separate option on top of that sequence. With it, the timestamp carries the presumption under Article 41(2) that the date and time are accurate and the data intact, which shifts the burden to the party disputing it. A blockchain anchor on its own does not carry that presumption, a distinction set out in blockchain timestamps compared with qualified timestamps.
Every certificate can be checked without us, using SHA-256, OpenSSL, the European Commission's trusted lists and any OpenTimestamps client. Our timestamp validator is open source and checks any provider's timestamp, not only ours.
Record your first deposit →Specifically enough that a court can tell what was covered, without setting out the content in a document the other side keeps. Drafters are caught between two failures: a definition covering everything disclosed may be treated as unreasonable and difficult to enforce, while a narrow list may leave out the material that actually mattered. One way through is to define the scope by reference to the documents themselves. Each document is hashed, the hash is timestamped, and the resulting certificate reference is listed in a schedule to the agreement. The schedule names a fingerprint rather than describing content, so the scope is exact and the material stays private.
It can be. Courts have been reluctant to enforce confidentiality definitions so wide that compliance becomes impractical or the restriction reaches beyond a legitimate business interest. Separately from enforceability, the breadth of a definition creates a pleading problem: a party bringing a claim has to particularise the information it actually relies on, and a clause covering everything gives no help with that. In IlliquidX Ltd v Altana Wealth Ltd and others [2024] EWHC 2191 (Ch), an application to amend particulars of claim, the court restated that claims for breach of confidence and misuse of trade secrets must be properly particularised as to both the information relied on and the misuse alleged, and refused some of the proposed amendments. IlliquidX went on to win at trial and on appeal, so this is not a story about a claim failing. It is about what the pleading has to contain first.
Each document is deposited before it is shared, which produces a certificate carrying the file name, its SHA-256 fingerprint, a timestamp from an accredited trust service provider and a blockchain anchor. The certificate has its own reference. That reference goes into a schedule to the agreement, so the confidentiality obligation attaches to identified documents rather than to a description of them. If the parties later disagree about whether a particular file was covered, the fingerprint answers it: the file either produces the listed fingerprint or it does not. This is a drafting approach rather than a legal rule, so the wording should be settled with the lawyer drafting the agreement.
Each version gets its own deposit and its own certificate reference, and the schedule is updated or extended as the negotiation proceeds. This is the situation static drafting handles worst, since a business plan revised three times or a model recalculated after every meeting leaves the parties arguing about which iteration was covered. A sequence of dated certificates removes that argument without requiring the agreement to be renegotiated each time.
Defining the scope and proving the disclosure are two separate problems, and a certificate answers only the first. A sent folder is held by the party relying on it, and a recipient download log is held by the opposing party, so neither carries weight on its own. Sharing from an access-controlled vault records each recipient and each view against a timestamp, and issuing individually watermarked copies means a leaked document identifies who received it. A scope definition without a disclosure record leaves half the question open.
More than many drafters assume, and this is worth knowing before relying on one. In the UK an NDA cannot be enforced to stop someone who is, or reasonably believes they are, a victim of crime from disclosing to the police or other bodies investigating or prosecuting crime, to a qualified lawyer for legal advice about it, or to a regulated professional for support. Agreements that attempt it are unenforceable to that extent. Confidentiality provisions also cannot be used to conceal unlawful behaviour, and the law here has been moving. None of that affects the ordinary commercial use of an NDA, but a confidentiality regime built on the assumption that an NDA silences everything is built wrong.
The same reasoning applies wherever an agreement refers to material held outside it. Licence agreements referring to licensed technology, development agreements referring to a specification, supply agreements referring to a bill of materials, employment contracts referring to defined confidential material. In each case the agreement describes something in prose while the actual subject sits in a separate file. Recording those files and referring to them by fingerprint closes the gap between what the contract says and what it points at.
It depends on where the dispute is heard. In the EU a qualified electronic timestamp carries a presumption under Article 41(2) of Regulation 910/2014 that the date and time are accurate and the data intact. The presumption is rebuttable, shifting the burden to the party disputing it. It applies across all EU member states, and the EEA states recognise qualified timestamps through the EEA Agreement. The UK retains an equivalent framework including the presumption, and recognition between the two regimes now runs one way: UK law continues to recognise EU qualified trust service providers, while the EU no longer recognises UK-registered ones. The United States has no equivalent statutory presumption, so the record is assessed under ordinary rules of evidence.
It does not make an unenforceable agreement enforceable, and it does not decide whether particular information qualifies as confidential. Those are questions of contract law and fact for a court. What it establishes is that a specific document existed in a specific form on a specific date and has not changed since, which is the foundation the other arguments are built on rather than a replacement for them. Drafting remains a job for a lawyer.
This page gives general information about confidentiality agreements and evidence. It is not legal advice, and nothing here is a recommendation on the drafting of your agreement. For a specific matter, consult a qualified lawyer in the relevant jurisdiction.