AI Contract Drafting for NDAs – Thailand & Vietnam 2026

When a customer says “send us an NDA by next week,” the first person to move at a Japanese-affiliated factory in Thailand or Vietnam is rarely a lawyer. It is an administration manager or the plant manager. Routing the request to the legal team at headquarters in Japan takes anywhere from a few days to a week; engaging a local law firm costs money. That is the moment AI contract drafting starts to look like a serious option. But a word of caution first. Most of what is written on this subject is really about review — running an already-drafted contract past an AI to check it — which is a fundamentally different job from drafting, where you build clauses from a blank page. NDAs carry an additional trap of their own: the very act of using AI can collide with the purpose of the document. This article starts from that distinction and works through what drafting with AI actually looks like in practice.

What “using AI for contracts and NDAs” actually means

Drafting and review are two different jobs

People say “we’re using AI on contracts” to describe two quite different pieces of work.

Drafting starts when no document exists yet. You translate the commercial reality into language, choose which clauses are needed, fill in the issues nobody thought of, and get the clause numbering and cross-references straight. It is the work of building a structure from nothing.

Review starts when a document is already in your hands — a contract sent over by the counterparty, or an existing in-house template. You are checking whether any clause works against you, whether the term and liability caps are reasonable, whether anything departs from internal standards. The decisive difference is that you begin with a document.

The two ask different things of an AI, and they fail in different ways.

Line up the failure modes and the danger in drafting becomes clear. Review fails by missing something that is there, and a second pair of eyes later on has a fair chance of catching it. Drafting fails by not noticing something that isn’t there. A contract with no governing-law clause reads perfectly naturally. The problem only surfaces when a dispute does.

We have already covered the review side elsewhere

Two other articles on this site deal with the review half of the job — checking existing contracts with AI. For tool categories, how to think about accuracy, and the issues worth checking, see AI Contract Review Guide 2026. For the organisational setup and operating design needed when you roll it out on a manufacturing site, see Introducing AI Contract Review in Manufacturing 2026. This article deals with the stage before either of those, when the document does not yet exist. In practice drafting and review are consecutive steps, so the second half of this article sets out a workflow that joins them together.

There is also common ground with How to Generate Proposal Documents with AI, in the sense of producing a structured document that will go outside the company. A fixed skeleton, contents that change per deal, and a human who takes responsibility for what goes out the door — that structure is identical in both cases. The difference is that in a contract, a small change of wording changes the legal effect.

Why AI-generated first drafts are spreading to manufacturing sites

First-draft time drops by an order of magnitude

Sirion, a contract management platform vendor, reports that for standard agreements such as NDAs, service agreements and employment contracts, producing a first draft — work that used to take roughly two to four hours — can be brought down to under five minutes. It is worth being precise about what is being compressed here: the time to a first draft, not the whole process through to signature.

Even so, the difference matters on the ground. Japanese-affiliated factories in Thailand and Vietnam frequently have no dedicated legal staff at the site, which leaves two options for a single NDA: join the queue at headquarters legal, or send it to a local firm. If a first draft can be stood up in minutes, at least the material needed to discuss internally — what actually has to be decided — is available the same day. Getting the discussion material out early shortens the number of days to signature.

The question is not “can AI write it” but “where do you put it”

An explainer published by the American Bar Association on generative AI in contract drafting frames this well. The question to ask is not whether an AI can write a clause, but where to position AI across a chain of tasks: first-draft generation, offering clause variants, preparing fallback language, issue spotting, and consistency checking. It is a placement problem.

That framing is practically useful. Treat “contract drafting” as one indivisible block and you are stuck with a binary — either you hand it to the AI or you don’t. Break it into steps and the parts you can delegate separate cleanly from the parts you cannot.

AI tends to work reasonably well on tasks like these.

• First-draft generation. Describe the deal and have it produce a draft in a standard structure
• Clause variants. Ask for the same confidentiality obligation in several forms — a three-year term and a five-year term, a narrowly scoped definition and a broad one
• Fallback language. Prepare a second and third position in advance, for when the counterparty rejects your first choice
• Issue spotting. Surface the clauses that would normally appear in this type of deal but are missing
• Consistency checking. Confirm that defined terms are used the same way throughout and that cross-references have not drifted

The tasks that should not be delegated are equally clear.

• The final judgement on legality. Whether a clause is valid under local law belongs to a lawyer
• Negotiation strategy informed by relative bargaining power. An AI has nothing to go on when deciding whether you can push or should concede
• Verification of commercial facts. Who delivers what, when it is inspected and accepted, who bears responsibility
• Deciding the level of risk the company will accept. Where to set the liability cap is a management decision

Roll AI out without sharing that dividing line internally and you end up with the worst possible operating pattern: an AI-generated draft going straight into the internal approval flow. Drawing the line on the process map — AI up to here, humans from here — is the real first step of adoption.

The documents manufacturing sites need most

These are the documents drafted most often at Japanese-affiliated factories.

• Non-disclosure agreements ahead of a new relationship, particularly when evaluating a new supplier
• Master purchase and sale agreements, the umbrella terms that individual orders sit under
• Service and outsourcing agreements, covering equipment maintenance, system development, on-site contracting and similar work
• Memoranda and minute-style agreements, used to amend part of an existing contract
• The standard terms printed on the reverse of purchase orders and order acknowledgements

Of these, NDAs are signed most frequently and carry the smallest financial weight per document, which is exactly why careful consideration tends to get skipped. And that is precisely what leads to the problem described next.

The overlooked paradox of NDAs and AI

Drafting a document that protects secrets, by handing the secrets over

The purpose of an NDA is to protect the confidential information you receive from the other side. Yet the moment you ask an AI to draft one, the prompt usually looks something like this.

“I need an NDA with Company A for discussions on joint development of automotive resin components manufactured at our Plant B. The discussion period is six months, and the information to be disclosed includes mould design data and defect-rate figures.”

That single sentence contains the counterparty’s name, your product field, your site, the purpose of the deal, and the categories of information you plan to disclose. The harder you try to improve the quality of the contract, the more specific the input becomes. This is the NDA-AI paradox: in the process of producing a document designed to protect a secret, you hand that secret to an outside service.

A commentary published by Bloomberg Law argues that this should be treated as a new baseline condition of NDA drafting. With consumer-oriented AI tools — especially free plans and individual subscriptions — whatever you type is processed on the provider’s servers and, depending on how the terms of service are written, may be used as model training data. In that case the act of entering the information can itself run up against the NDA’s own restriction on who confidential information may be disclosed to.

Why return-or-destroy clauses become technically impossible to honour

More awkward still is the collision with the return-or-destroy obligation found in most NDAs. On termination, or on the counterparty’s demand, you must return or destroy all confidential information received along with every copy, and certify in writing that you have done so. This clause is present in virtually every Japanese corporate template.

The question is whether that obligation can be technically performed for information typed into an AI tool. As the contract-focused legal blog terms.law sets out, the most a user can do is delete the chat history, and that does not mean the information is gone on the provider’s side. In reality it may persist in places like these.

• Server logs on the provider’s side, recording who sent what and when
• Cached prompts, held temporarily for processing efficiency
• Datasets used to retrain the model — once training has happened, the content cannot be extracted and deleted
• Backups, taken on a schedule and retained for a set period
• Sub-processor systems, where the provider outsources infrastructure or part of the processing to another company

In other words, even after the history disappears from your screen, you can no longer honestly certify to the counterparty that the information was destroyed. Not being able to certify means not being able to rebut an allegation that you breached the contract. Return-or-destroy clauses are not formalities; they are among the clauses actually invoked when things go wrong.

Another point that gets overlooked is the scope of permitted recipients defined in the NDA. Most templates limit disclosure to “officers and employees who need to know for the Purpose” and require prior written consent before any disclosure to an external contractor. An external AI service is hard to fit into any part of that definition. It is not an officer, not an employee, and not a pre-approved contractor. Depending on how the clause is worded, a reading emerges under which the input constitutes disclosure to a third party without consent. How this is actually interpreted turns on the contract language and on local law, so if you are operating on an existing template, we recommend having a lawyer confirm this point.

What to do about it in practice

This problem does not have to be avoided by refusing to use AI. It can be designed around.

First, you can address it in the contract language itself. Give AI tool inputs an explicit position in the NDA’s definition of confidential information and in its permitted-recipients clause. One approach is to include generative AI services used by either party in their business within the scope of permitted recipients, on condition that they are configured so that inputs are not used for training. Agreeing this with the counterparty up front leaves much less room to argue over interpretation later. How such a clause should be drafted depends on the structure of the contract as a whole and on how local law treats it, so always have a lawyer check the wording.

Second, choose the tool. Some enterprise AI services contractually commit that inputs will not be used to train models. Check that this “no-training” condition exists as a term of the contract or the terms of service, rather than as a line in a sales deck. Confirm in writing, at the same time, where the data is processed and stored, how long it is retained, and whether deletion requests can be honoured. If the tool will be used at a site in Thailand or Vietnam and personal information may be involved, Thailand’s PDPA and Vietnam’s personal data protection rules come into the picture as well.

Third, change what you put in. This is the most reliable measure. Replace concrete details — the counterparty’s name, amounts, quantities, product names, site names, delivery dates — with placeholders before anything goes into the AI.

• Replace the counterparty’s name with “the Disclosing Party” / “the Receiving Party” or “[Counterparty Name]”
• Abstract the product name to “the Product” and the site name to “the Plant”
• Turn amounts, quantities and periods into placeholders such as “[Amount]” and “[Quantity]”
• Reinsert the real values into the generated draft inside a secure internal environment

Contract clauses are not written to depend on proper nouns in the first place. Given that “the Disclosing Party shall provide the Receiving Party with…” works perfectly well, there is in fact almost never a need to give an AI the real names. Making that one extra step a standard procedure resolves most of the paradox.

These three measures are not alternatives; they should be layered. Even with rigorous placeholder discipline, a tool with a contractual no-training commitment is still the safer choice.

The legal standing of contracts and e-signature in Thailand and Vietnam

Thailand’s law as it stands (Electronic Transactions Act, B.E. 2544)

In Thailand, the Electronic Transactions Act, B.E. 2544 (2001) gives electronic signatures the same legal effect as a signature on paper. This is already settled under current law; a contract concluded electronically is not somehow ineffective.

There are exclusions, however. Documents concerning family and succession matters such as wills, and documents requiring registration or filing with a government authority such as property registration, sit outside the framework. In factory operations, the safe assumption is that for anything touching land and buildings, and for filings with the authorities, electronic execution alone may not be sufficient. Which documents actually fall outside the scope has to be checked case by case, so for any procedure involving a government filing, consult local counsel or a legal specialist in advance.

There is also a practice specific to Thailand: alongside the signature, an electronic company seal is frequently expected. This is closer to commercial custom than a statutory requirement, but when the counterparty is a local company, documents genuinely do get rejected for want of a company seal.

Separately, where a signature meets the requirements of a “reliable electronic signature,” its validity is presumed as a matter of law, and the burden of proof shifts to the party asserting invalidity. How strict to be about signature method is a judgement call driven by the importance of the contract.

Thailand’s 2026 draft amendment has NOT been enacted

This point needs handling with care. Thailand’s Electronic Transactions Development Agency (ETDA) put a full overhaul of the Electronic Transactions Act out for public hearing from May 12, 2026 to June 15, 2026. The draft includes directions that could affect day-to-day practice.

• Expressly bringing biometric data within the definition of an electronic signature
• Recognising the binding force and validity of contracts formed between automated systems, or between a system and an individual, without human visual confirmation or intervention
• Introducing protection allowing unintended input errors to be withdrawn, provided notice is given promptly after discovery

The second of these could be a significant change for companies contemplating operations in which orders or contracts are formed automatically between systems.

However, as of August 2026 this draft has not been enacted as law. It still has to go through parliamentary review, and enactment and entry into force are expected to take roughly a further year from now. For the time being, then, the appropriate posture is to build your practice on the Electronic Transactions Act as it stands (B.E. 2544) and to track the amendment as information only. Designing operations around the draft’s contents is premature, and the final application will need to be confirmed with a lawyer against the enacted text and implementing regulations.

Vietnam’s Law on Electronic Transactions 2023

In Vietnam, the National Assembly passed the revised Law on Electronic Transactions (Law 20/2023/QH15) on June 22, 2023, and it took effect on July 1, 2024. It replaces the earlier law enacted in 2005 and — unlike the Thai draft — is already in force.

Two points matter in practice.

The first is the originality of data messages. Where the integrity of the content has been preserved from the moment it was created, and it is stored in a form that can be accessed and used when needed, the data message has the same effect as an original. Put the other way round: if your storage method cannot demonstrate that the content has not been altered, asserting it as an original becomes difficult.

The second is the treatment of electronic signatures and certificates issued by foreign certification providers. These are legally recognised once the procedure set by the competent ministry (the Ministry of Information and Communications at the time of enactment) has been completed. If you want to use the same electronic signature infrastructure at your Vietnamese entity as at headquarters in Japan, check before filing whether this procedure applies and which body currently handles it, given subsequent ministerial reorganisation.

Thailand and Vietnam compared

This table is a general summary as of August 2026, and the treatment changes depending on the contract type and on whether a government filing is involved. Confirm actual applicability with local counsel.

“Drafting with AI” and “a contract being validly formed” are separate questions

This is the most important distinction in the article. Everything above about electronic transactions law concerns the execution process — how a contract is formed and how it is signed. That is an entirely separate question from an AI writing the text of a draft.

The same applies to the “validity of contracts formed between automated systems” in Thailand’s 2026 draft amendment. That provision contemplates situations where a contract is concluded without human involvement. It has nothing to do with an AI writing the draft of a contract whose content a human ultimately reviews and signs. Explanations that conflate the two are not hard to find, so this is worth being careful about.

Stated plainly: neither Thai law as it stands nor Vietnamese law as it stands places any particular restriction on who — or what — prepares the text of a contract. What determines the effect of a contract is whether the parties intended to enter into it, agreed on its content, and signed by a valid method. Drafting with AI is therefore, in itself, unlikely to be problematic under current law — or under the draft amendment described above, were it to be enacted — provided the final execution is carried out properly. Judgements about any specific contract presuppose confirmation with a lawyer.

What can become a problem is a breach of confidentiality or data protection obligations during the drafting process, and defects in the content of the clauses generated. The first is the issue covered in the previous section; the second is the operational issue covered next.

A workable workflow — AI drafts, humans decide

Build the templates and approved clause library first

Having an AI write a contract from a blank page is, in fact, the least efficient and highest-risk way to use it. The clauses it generates may be sound as general propositions, but they will not match the wording your legal team has previously signed off. The result is that you have to check the entire content from scratch every time.

The first thing to build is your internal templates and an approved clause library. Pull the clauses that have been through legal or outside counsel from contracts you have already signed, and organise them by type. For the major issues — scope of confidentiality, term, return-or-destroy, governing law, dispute resolution, liability caps — hold both a standard position and a fallback position.

With that in place, assign the AI roles like these.

• Work out the list of clauses required, from a description of the deal
• Select the applicable clauses from the library and assemble a draft with the deal information filled in
• Offer several clause options for issues the library does not cover
• Check the consistency of defined terms and the coherence of article numbering in the finished draft

Structured this way, the proportion of genuinely new text the AI has produced falls, and the range a human needs to check becomes explicit. The workload drops not because the AI writes prose, but because you can identify where the checking has to happen.

Keep specifics as placeholders throughout

As set out earlier, the counterparty’s confidential information and the commercial terms should be converted to placeholders before anything reaches the AI. In practice the following routine works.

• Create a “fact sheet” for each deal in an internal system, recording company names, amounts, quantities and dates there
• Give the AI only the abstracted conditions and have it build the draft
• Merge the fact-sheet values into the finished draft internally
• Never re-submit the merged, complete version to an external AI service

That last point is the one most easily missed. However rigorous your placeholder discipline is at drafting time, throwing the completed version at an AI with “give this a final check” hands over the specifics right there. Apply the same standard at both the drafting and the checking stage.

Generated clauses always need a final legal review

Drafts produced by AI require expert confirmation. The following clauses in particular need to be checked individually, because errors only surface once a dispute has begun.

• Governing law. Thai law, Japanese law, or the law of a third country. Is it appropriate given the counterparty’s location and the place of performance
• Dispute resolution. Court jurisdiction or arbitration. If arbitration, which institution, in what language, seated where
• Limitation of liability. The cap amount, exclusion of indirect and consequential loss, and the carve-outs from the cap
• Confidentiality term and survival. The scope and duration of obligations that continue after termination
• Force majeure. What counts as force majeure and what follows if it occurs
• IP ownership. In joint development, which party owns the rights in the deliverables

Because plenty of specimen wording for these is in circulation, an AI will readily produce something that looks the part. Whether that wording is appropriate against your actual commercial arrangement and against local law is not something you can determine by reading the clause. This is where the money should be spent.

Multilingual contracts always need a governing language clause

In Thailand and Vietnam, contracts are frequently produced in more than one language — Japanese and English, English and Thai, or all three side by side.

In that situation the one thing you must include is a governing language clause, specifying which language version prevails if the versions diverge in interpretation. Without it, a dispute begins with an argument about which text you are even supposed to be interpreting.

If you are having an AI produce the multilingual versions, build the following into your process.

• Decide internally which language governs, before involving the AI at all
• Put the governing language clause into the template from the outset
• Finalise the governing-language version first, then produce the other versions. Do not run them in parallel
• Have a human verify translation accuracy in the other versions, especially figures, dates and units of currency
• For documents that must be filed with the authorities, confirm in advance whether a local-language version is required

Translation accuracy is an area where AI genuinely performs well, but contract translation starts from a design decision about which text is authoritative, and that is a legal judgement.

Three stages — drafting, AI review, human sign-off

Putting all of the above together, the workable workflow has three stages.

The first stage is drafting. Using the templates and the clause library as a base, have the AI assemble a draft from placeholder-only conditions. What you get is a document to start the discussion from, not a finished product.

The second stage is AI review. Check the draft you have produced from a different angle: missing issues, inconsistent defined terms, broken cross-references, departures from internal standards. How to run this step is covered in AI Contract Review Guide 2026. Checking along the same line of reasoning you used to draft will let things slip through, which is why it matters that review is done from a different angle and against different criteria.

The third stage is human sign-off. A legal team member, or outside counsel, confirms that the content is sound and lawful. This stage cannot be skipped. For how to embed this three-stage process at a manufacturing site and how to assign approval authority, see Introducing AI Contract Review in Manufacturing 2026.

The important thing is to decide, for each of the three stages, what record remains. Which version was produced by whom and when, what was changed, and who approved it. With that record you can trace, after the fact, why a clause ended up the way it did. A contract is not finished when it is signed; it is a document people read again years later.

How to avoid a failed rollout

Common failure patterns

When adoption goes badly, the cause is usually one of the following.

• The tool was rolled out without building the templates and clause library, so everything still has to be checked from scratch and no time is saved
• The approval process was skipped on the assumption that “the AI produced it, so it’s fine” — discovered years later, in a dispute
• The site was given access before any rules on entering confidential information were set, and by the time anyone noticed, several deals’ worth of information had gone to an external service
• Drafting and review were run through the same tool and the same procedure, reducing the check to a formality
• Confirming local law was left until later, and requirements around governing language or government filings surfaced late enough to force a rewrite

None of these are tool performance problems. They are operating design problems.

Where to start

For a first rollout, we recommend narrowing the scope to NDAs alone. There are three reasons. The volume is high, so the effect is visible. The structure is relatively standardised and easy to template. And working through the paradox described in this article forces you to settle the rules on entering confidential information before anything else.

Once NDAs are running smoothly, extend to master purchase agreements and service agreements. Taking them in that order makes it far less likely that you will have to tear up and rebuild the operating rules midway through.

Where TOMAS TECH fits

Let us be explicit about this. TOMAS TECH is not a law firm and we are not lawyers. Judgements on the legality of contract clauses, legal advice on contract content, and handling disputes are legal matters and fall outside our scope of work. Always take those to a lawyer or a legal specialist.

What we can support is the systems layer that sits before and around all that. Concretely: how to organise your internal templates and approved clauses and where to hold them in your systems; where deal information is managed and how it flows into the placeholders; how to embed the rules for AI tool inputs into your standard operating procedures and design a workflow specifying who approves at which stage; and how to support, in a system, an approval flow that runs between sites in Thailand and Vietnam and headquarters in Japan.

Legal substance to the specialists, the systems layer to us — that is the accurate division of labour.

Frequently asked questions

Is AI contract drafting illegal?

Neither Thailand’s Electronic Transactions Act (B.E. 2544) nor Vietnam’s Law on Electronic Transactions (Law 20/2023/QH15) contains any provision restricting who may prepare the text of a contract. What determines a contract’s effect is whether the parties intended to enter into it, agreed on its content, and signed by a valid method. So the general understanding as of August 2026 is that no provision prohibits an AI from producing a draft as such. If confidentiality or data protection obligations are breached during the drafting process, that is a separate matter. Rules on handling the legal affairs of others as a business also differ from country to country, so if you are considering offering contract drafting as a service to third parties, that requires separate analysis. In any event, treat confirmation with a lawyer or legal specialist as a precondition for any final judgement on whether your specific practice is acceptable.

Can we use ChatGPT to draft an NDA?

It depends how. Entering the counterparty’s name or specific commercial terms directly into a free plan or an individual subscription should be avoided. Inputs are processed and stored on the provider’s side and, depending on the terms of service, may be used for training — at which point you can no longer reconcile your practice with the NDA’s limits on permitted recipients or its return-or-destroy obligation. If you are going to use it, choose an enterprise plan with a contractual no-training commitment, and replace the counterparty’s specifics with placeholders before entering anything. Meet both of those conditions and the practical risk drops substantially. When in doubt, check against your internal information management rules and confirm with legal.

Can we use an AI-generated contract as it is?

We would not recommend using one as it is. The clauses an AI generates may look reasonable as generic specimen wording, but whether they are appropriate for your actual commercial arrangement and for local law is another question entirely. Governing law, dispute resolution, liability caps and IP ownership are particularly unforgiving: errors are invisible on reading and only surface once a dispute begins. And because drafting fails by omitting required clauses entirely, checking what is written on the page is not enough. Treat an AI draft as the starting point for discussion, and have the final content confirmed by a legal team member or a lawyer.

Is a contract drafted by AI valid in Thailand and Vietnam?

Validity is determined by the execution process, not by who wrote the draft. In Thailand, the Electronic Transactions Act (B.E. 2544) gives electronic signatures the same effect as written signatures, and in Vietnam the Law on Electronic Transactions in force since July 2024 gives electronic execution legal effect. So a contract drafted by an AI has the same effect as any other contract, provided the parties agree on the content and sign by a valid method. Note that although a full draft amendment to Thailand’s Electronic Transactions Act went to public hearing between May and June 2026, it had not been enacted as of August 2026 and the current law applies. Confirm specific contract types with local counsel.

Should we adopt drafting AI or review AI first?

It depends on your situation. If you are frequently assessing contracts sent to you by counterparties, starting with review will show results sooner. If you are more often the party presenting the contract — particularly issuing NDAs on your own template — starting with drafting fits your practice better. At Japanese-affiliated factories in Thailand and Vietnam, NDAs are often issued by your side during new supplier evaluation, so starting with drafting is a common pattern. Either way you will end up combining the two, so we recommend sketching the full three-stage picture — drafting, AI review, human sign-off — right at the start.

Summary

Drafting contracts and NDAs with AI is a different job from reviewing them. Review fails by missing something that is there; drafting fails by not noticing something that isn’t. That difference changes how you have to check.

On top of that, NDAs carry a paradox of their own. In the process of producing a document meant to protect a secret, you hand that secret to an outside service. Factor in the return-or-destroy obligation and it gets worse: deleting the chat history does not reach the provider’s logs and backups, so you can no longer certify that the information was destroyed. There are three responses — address it in the contract wording, choose a tool with a contractual no-training commitment, and convert specifics into placeholders. The third can be started today and has the largest effect.

On the legal side in Thailand and Vietnam, the starting point is the distinction between drafting with AI and a contract being validly formed. Thailand’s 2026 draft amendment concerns contracts formed between automated systems; it is not about AI-generated drafts. And that draft had not been enacted as of August 2026, so the current Electronic Transactions Act (B.E. 2544) applies. Vietnam’s Law on Electronic Transactions 2023 took effect in July 2024 and is already in force. In both cases, how the rules apply in your specific situation is something to confirm with a lawyer.

Operationally, build the templates and approved clause library first, limit the AI to filling in blanks and proposing candidate clauses, and run the three stages of drafting, AI review and human sign-off. That is the realistic shape of it.

TOMAS TECH is not a law firm, so we leave judgements on the legality of clauses and on contract content to lawyers. What we can help with is the systems side: how to organise templates and clause libraries within your systems, where to manage deal information and how to feed it into placeholders, how to build AI input rules into your standard procedures, and how to design an approval flow that spans your sites and your Japanese headquarters. Early exploratory conversations are perfectly welcome — before you have chosen a tool, or while you are still working out where to begin. We are happy to start by walking through your current document management and approval flow and working out together what to put in place first. Get in touch through our contact page.

References

• NDA Automation and Tracking Playbook — Sirion — the statement that for standard agreements, first-draft preparation that previously took roughly two to four hours can be reduced to under five minutes
• Using Generative AI in Contract Drafting — American Bar Association — the framing that the question is where to position AI rather than whether it can write clauses, and the breakdown into first-draft generation, clause variants, fallback language, issue spotting and consistency checking
• Non-Disclosure Agreement Drafting Must Account for AI’s Risks — Bloomberg Law — the point that entering information into consumer-oriented AI tools can conflict with an NDA’s permitted-recipients clause
• AI in NDAs — How to Stop Your Secrets from Becoming Training Data — the problem that deleting chat history does not reach server logs, caches, retraining datasets, backups and sub-processor systems, leaving the return-or-destroy obligation impossible to certify
• Electronic Signatures and Electronic Contracts in Thailand — BUSINESS LAWYERS — the effect of electronic signatures under Thailand’s Electronic Transactions Act (B.E. 2544), the scope of exclusions, the electronic company seal in practice, and the presumption of validity for a reliable electronic signature
• Thailand Set to Overhaul Its E-Transactions Framework — Tilleke & Gibbins — the ETDA public hearing held from May 12 to June 15, 2026, the treatment of biometric data, the validity of contracts formed between automated systems, and the outlook for enactment and entry into force
• Unpacking Vietnam’s Law on E-Transactions 2023 — International Bar Association — the passage of the law on June 22, 2023 and its entry into force on July 1, 2024, the originality requirements for data messages, and recognition of foreign providers’ electronic signatures and certificates once the procedure set by the competent ministry (the Ministry of Information and Communications at the time of enactment) is completed