Using AI to Draft Your NDA? 10 Legal Risks Every Business Should Know Before You Sign

A professionally written NDA does not always mean a professionally protected business.

Just a few years ago, it was common for clients, who downloaded free Non-Disclosure Agreement (NDA) templates from the internet, to come to us for an NDA review. They would make a few amendments themselves before asking us to review the document.

Today, we are seeing a noticeable shift.

More businesses are preparing their own NDAs using AI tools such as ChatGPT, Claude or Gemini before sending the draft to us for legal review. In many cases, the document is surprisingly well written. The language is polished. The clauses appear comprehensive. At first glance, the agreement looks like it was prepared by an experienced lawyer.

As corporate and intellectual property lawyers, we welcome this development.

Artificial intelligence has become an excellent drafting assistant. It can produce a first draft within minutes. It can improve efficiency. It can even suggest clauses that many business owners may not have thought about.

However, after reviewing a growing number of AI-generated documents, we have noticed a recurring pattern.

The problem is rarely poor English or missing clauses (sometimes too many clauses).

Instead, the real issue usually lies beneath the surface.

The agreement may fail to reflect the commercial arrangement between the parties. It may allocate legal risks unintentionally. Different clauses may point in different directions. Worse still, the document may create uncertainty over confidential information, IP ownership or future rights that neither party intended.

These are not drafting problems. They are legal judgement problems.

One question often guides our review:

If this business relationship breaks down five years from now, how will a judge interpret this agreement?

That question changes how lawyers read contracts. It also explains why a document that seems professionally drafted can still expose a business to unnecessary legal risk.

In this article, we share some of the issues we encountered when reviewing AI-generated NDAs and free NDA templates.

A Real-Life Example: When a Good-Looking NDA Was Not Good Enough

Recently, we advised a software developer firm that prepared its own NDA using ChatGPT.

On first reading, the draft NDA looked impressive.

It contained detailed definitions, lengthy confidentiality obligations, IP provisions, improvements, publicity, governing law, dispute resolution, residual knowledge and other clauses that are commonly found in commercial agreements.

“Does this look alright?” was the question we received. Our answer was more complicated than a simple yes or no.

Individually, many of the clauses were reasonable. Collectively, they created uncertainty.

For example, some clauses suggested that all intellectual property developed during the relationship would belong to one party. Other provisions implied that joint ownership should be considered.

We made several revisions before the agreement properly aligned with the client’s objectives.

This shows an important point.

The AI-generated NDA may look okay. The real challenge is whether the agreement functions as a coherent legal framework.

Problem 1: AI Understands Clauses. It Does Not Necessarily Understand the Commercial Deal.

AI generally produces agreements by predicting what each clause should say based on similar documents it has seen before.

Each clause may appear reasonable when viewed independently. The problem arises when they are read together; they may create ambiguity.

We observed this repeatedly when reviewing one AI-generated NDA.

Problem 2: AI Is Only as Good as the Prompt. The Real Challenge Is Knowing What Prompt to Write.

The quality of an AI-generated NDA depends heavily on the prompt entered. However, the real challenge is often not writing a better prompt. It is knowing what legal or commercial points should be included in that prompt.

A detailed prompt will almost always produce a better draft than a vague one. The user can specify the governing law, identify the parties, describe the transaction, explain the commercial objectives, and even ask AI to include particular legal provisions.

In many cases, this produces a much stronger first draft.

However, legal drafting presents a different challenge.

The real question is not whether you know how to write a better prompt.

It is whether you know enough about the transaction and the law to know what should be included in that prompt in the first place.

AI can only respond to the information and instructions it receives. It cannot reliably identify legal issues that the user never thought to raise.

For example, imagine a software developer asks ChatGPT:

“Draft an NDA governed by Malaysian law that protects my interests as a software developer.”

This is probably one of the most common prompts a business owner would use.

ChatGPT will likely produce a professionally drafted agreement. It may contain detailed confidentiality obligations, intellectual property clauses, and other standard legal terms.

At first glance, the document may appear comprehensive. However, the AI has no information about the issues that matter most unless the user specifically asks.

For example:

• Does the software developer already own proprietary source code, software frameworks or reusable components that should remain its Background IP?

• Which issues should be addressed in the Software Development Agreement instead of the NDA?

  
  

Ultimately, the value of AI in legal drafting begins with better judgement about what should be included, omitted, and what legal issues must be addressed before the first prompt is ever written.

AI can draft the agreement you ask for. It cannot reliably identify the agreement you actually need.

Problem 3: AI Often Confuses Confidentiality with Intellectual Property Ownership

Keeping information confidential is not the same as owning the intellectual property behind that information. This is one of the most common misconceptions we encounter.

Many business owners assume that once an NDA has been signed, ownership of their intellectual property is automatically protected.

That is not how an NDA works. An NDA primarily regulates how confidential information may be used, disclosed or protected.

It does not automatically determine who owns patents, source code, product designs, or other IP created before or during the relationship.

These are separate legal issues.

Unfortunately, AI-generated NDAs may blur this distinction. The document may introduce ownership concepts into confidentiality provisions. This creates unnecessary uncertainty.

Many businesses believe: “If we have signed an NDA, our intellectual property is protected.”

A more accurate statement would be:

“An NDA protects confidential information. Intellectual property ownership often requires separate contractual provisions.”

Understanding this distinction can prevent expensive disputes later.

Problem 4: More Clauses Do Not Always Mean Better Protection

Many people often assume that a long agreement is a better agreement. In reality, unnecessary clauses can introduce uncertainty, complicate negotiations, or even create obligations that neither party intended.

One of the biggest differences between experienced legal drafting and AI drafting is knowing what to leave out.

When AI is asked to prepare an NDA, it often searches for clauses that commonly appear in similar agreements. The result is a document packed with provisions covering almost every possible legal issue.

We frequently see clauses dealing with inventions, improvements, non-competition, moral rights, alternate dispute resolution & etc.

None of these clauses is necessarily wrong. The real question is whether they are appropriate for the particular transaction.

An NDA should be drafted around the commercial relationship, not around every clause that AI believes might be relevant.

For example, an SME was discussing a potential collaboration.

The objective was simple.

One party wanted to disclose confidential production know-how to allow preliminary discussions.

The AI-generated NDA, however, included many unnecessary clauses and issues.

None of these issues had been discussed commercially. The parties ended up negotiating matters they had never intended to negotiate. Ironically, the additional clauses delayed the transaction instead of protecting it.

Problem 5: AI Cannot Replace Commercial Legal Judgement

Artificial intelligence is an excellent drafting assistant. It is not yet a substitute for legal judgement developed through experience advising businesses in real commercial transactions.

After reviewing numerous AI-generated NDAs, one conclusion has become increasingly clear.

The quality of AI drafting has improved dramatically.

However, the final 10 to 20 percent of legal review often determines whether the agreement truly protects the client’s interests.

That final stage involves questions AI still struggles to answer reliably.

For example:

• Does this agreement accurately reflect the commercial deal?

• Are the parties allocating legal risk intentionally?

• Does every clause support the same objective?

• Will these provisions assist or hinder future negotiations?

• If litigation arises years later, how will this wording likely be interpreted?

  
  

These questions require more than language generation. They require judgement developed through practical experience.

When Is a Free or AI-Generated NDA Probably Good Enough?

A free NDA template or an AI-generated NDA may be sufficient where the commercial risk is low or the discussion is at preliminary stage.

Not every NDA requires extensive legal drafting.

In many situations, a simple confidentiality agreement is sufficient because the commercial relationship has not yet developed to the stage where complex legal issues arise.

For example, a simple NDA may be suitable where the parties are:

• exploring a potential business collaboration for the first time;

• holding an introductory meeting to assess commercial interest.

  
  

In these situations, the parties' objective is simply wishing to discourage unauthorised disclosure while allowing discussions to proceed efficiently.

When Should You Definitely Obtain Legal Advice?

Legal review becomes significantly more valuable once confidential information is tied to intellectual property, technology development or long-term commercial collaboration.

As the commercial relationship becomes more sophisticated, so do the legal risks.

At this stage, an NDA is no longer just protecting confidential information. It often becomes the foundation of the coming future agreements, especially where the parties are entering into substantive commercial arrangements.

Legal advice should be seriously considered where the NDA relates transactions that raise questions that extend well beyond confidentiality, such as software development projects, franchising arrangements or manufacturing partnership.

What Businesses Should Consider Before Signing an NDA

Before signing any NDA, it is worth asking a few practical questions.

First, what confidential information will actually be protected?

Not all confidential information carries the same commercial value. Customer lists, software source code, proprietary manufacturing methods and trade secrets usually require a higher level of protection than general business information.

Secondly, does the NDA accurately reflect the commercial relationship?

An agreement prepared for a manufacturing project may not be suitable for a software development engagement. Likewise, an NDA used during preliminary discussions may not be appropriate once the parties begin jointly developing technology.

Finally, ask whether every clause genuinely serves a purpose.

A longer agreement does not necessarily provide greater protection. Every provision should support the parties’ common goals.

Frequently Asked Questions

Can I use ChatGPT to draft an NDA?

Yes. AI tools such as ChatGPT can be an efficient way to prepare a first draft, particularly for simple confidentiality arrangements. However, the draft should be reviewed carefully before signing, especially where valuable intellectual property, software or trade secrets are involved.

Is a free NDA legally valid in Malaysia?

A free NDA may be legally enforceable if it satisfies the general requirements of a valid contract under the Contract Act 1950. However, many free templates are drafted for generic situations. They may not adequately address the specific commercial risks of your transaction.

Can an AI-generated NDA be relied upon without legal review?

That depends on the transaction. For low-risk preliminary discussions, it may be sufficient.

Where valuable confidential information or intellectual property is involved, legal review is usually worthwhile because the greatest risks often arise from commercial issues rather than drafting quality.

Should ownership of software developed during a project be included in an NDA?

Generally, ownership of software, deliverables and intellectual property is better addressed in the Software Development Agreement or another principal commercial agreement.

Attempting to resolve these issues within an NDA may create uncertainty if the commercial terms have not yet been fully negotiated.

Why do lawyers make so many amendments to an AI-generated NDA when it already looks professionally drafted?

Because legal review is rarely about correcting grammar.

It is about ensuring the agreement accurately reflects the commercial arrangement, allocates legal risk appropriately and remains internally consistent.

Many amendments involve judgement rather than drafting.

Need a Second Opinion on Your AI-Generated NDA?

AI can help you prepare a first draft in minutes. The more important question is whether that draft accurately protects your confidential information, intellectual property and commercial interests.

At LAWENCO, we regularly advise businesses on reviewing, drafting and negotiating NDAs, Software Development Agreements, Licensing Agreements, Personal Data Protection Notice, Privacy Notice and other commercial contracts. Our focus is not simply on improving the wording of an agreement, but on ensuring it reflects the commercial deal, allocates legal risk appropriately and protects your business in the long term.

If you have prepared an NDA using AI or a free template and would like an experienced legal review before signing, we would be pleased to assist!

Written by,

Lawrence Tan 

Registered Trademark, Patent and Design Agent

LL.B (HONS), CLP

Advocate & Solicitor

Disclaimer

The discussion in this article is intended for general information only. It should not be regarded as legal advice, nor should it be relied upon as a substitute for obtaining advice on your specific circumstances. Every business relationship, transaction and NDA should be assessed based on its own commercial objectives and legal considerations.

The observations in this article are also based on our present experience reviewing AI-generated NDAs. Artificial intelligence is developing rapidly, and its drafting capabilities continue to improve. Some of the limitations discussed above may become less significant or even be addressed as AI technology evolves.