AI Can Draft a Contract. But Can It Protect Your Business?

Generative AI can produce a contract in seconds.

It can create something that looks polished, uses legal terminology and includes many of the clauses you would expect to see in a commercial agreement.

That can make it tempting to think that most of the work has already been done.

But there is an important difference between generating a legal document and creating an agreement that properly protects a business.

When I look at AI-generated agreements, the issue is usually not whether the document looks professional. Many of them do.

The harder questions are:

• Does the agreement actually reflect the commercial deal?

• Has it identified the risks that matter to the business?

• Has it allocated those risks appropriately?

• Does it comply with the laws that apply?

• Do the clauses work together?

• And, perhaps most importantly, what is missing?

That last question can be particularly difficult.

A contract can be 20 or 30 pages long, look comprehensive and still leave important legal and commercial risks completely unaddressed.

For significant commercial arrangements, the value of a well-drafted contract is therefore not simply the words on the page.

It is the legal and commercial thinking behind them.

AI can generate clauses. It does not necessarily understand your deal.

A contract should be built around what is actually happening between the parties.

That sounds obvious, but it is one of the most important parts of getting an agreement right.

Before drafting a contract, I would usually want to understand things such as:

• What is each party actually providing?

• What are the important deliverables?

• Who is responsible for different parts of the service?

• How and when will payment occur?

• What intellectual property is being created or used?

• What information will be shared?

• What happens if the service does not work as expected?

• What happens if someone wants to leave the arrangement?

Those questions determine what needs to go into the agreement.

AI can produce clauses about payment, intellectual property, confidentiality, warranties, indemnities and termination.

But the presence of those clauses does not necessarily mean the agreement has arrived at the right position for the transaction.

The question is not simply:

“Does the agreement contain an IP clause?”

It is:

“Does this IP clause correctly deal with the IP involved in this particular transaction?”

The same applies to almost every important contractual provision.

The biggest risk can be what is missing

This is one of the limitations that business owners may not immediately see.

An obviously badly drafted clause is relatively easy to identify.

A missing clause is much harder.

If you do not work with contracts regularly, you may not know that something should be there in the first place.

I frequently see agreements that appear comprehensive but do not fully address the risks arising from the commercial arrangement.

That may include gaps around:

• intellectual property;

• data use;

• confidentiality;

• subcontracting;

• payment;

• service failures;

• termination;

• post-termination obligations;

• liability; or

• regulatory compliance.

The appropriate protections will depend on the particular business and transaction.

That is why the number of pages in an agreement is not necessarily a good indication of how well protected a business is.

A longer contract is not automatically a better contract.

Legal wording can sound convincing and still be wrong

Generative AI can also produce inaccurate legal information.

One of the difficulties is that the wording can sound authoritative even where the underlying legal position is incorrect.

I see issues such as:

• incorrect references to legislation;

• legal concepts that do not apply to the transaction;

• provisions that do not reflect Australian law;

• clauses that may not comply with applicable regulatory requirements; and

• contractual positions that create unnecessary legal risk.

This can be particularly important when dealing with areas such as privacy, consumer law, unfair contract terms and intellectual property.

A clause can look completely standard and still create a problem.

For example, a provision may attempt to give one party very broad rights or impose significant obligations on the other party. Even if the wording looks like something commonly found in a contract, it still needs to be considered against the laws that apply and the circumstances of the transaction.

The same applies to references to legislation.

If an agreement refers to a particular Act or legal requirement, that reference should be checked rather than assumed to be correct simply because the language sounds legal.

A contract needs to work as one document

Another common issue is that individual clauses may look reasonable when read separately but do not work properly when read together.

For example, one clause may say payment is due within 14 days while another gives the customer 30 days to dispute an invoice.

A termination clause may say certain rights end immediately, while another clause appears to allow those rights to continue.

An indemnity may appear to be subject to a liability cap in one section but fall outside it in another.

These are not always obvious drafting errors.

The clauses themselves may each look perfectly acceptable.

The problem is the way they interact.

A contract needs to operate as one coherent document.

That requires more than producing a collection of standard clauses.

Intellectual property is a good example of why context matters

This is particularly relevant for technology, software, consulting and creative businesses.

It is common to see an agreement simply state that one party “owns the intellectual property”.

But the real position may be considerably more complicated.

You may need to distinguish between:

• intellectual property a party owned before the relationship started;

• intellectual property created during the engagement;

• reusable tools, methodologies, software or templates;

• customer-specific deliverables;

• third-party intellectual property;

• licences required to use the deliverables; and

• rights that need to continue after the agreement ends.

If AI is part of the product or service, there may also be additional questions around inputs, outputs, underlying technology, training data and permitted use.

A generic intellectual property clause may therefore look adequate while failing to reflect how the business actually operates.

That can become a serious issue later if the parties disagree about who owns something or what they are allowed to do with it.

The same applies to data and confidential information

Technology businesses in particular often have contractual arrangements involving significant amounts of data.

A basic confidentiality clause may not be enough.

Depending on the arrangement, the agreement may need to deal with questions such as:

• what information can be used;

• what it can be used for;

• whether it can be disclosed to subcontractors;

• how it needs to be protected;

• whether it can be retained after the relationship ends;

• whether it needs to be returned or deleted;

• whether personal information is involved; and

• whether additional privacy or security obligations apply.

Again, the legal drafting needs to match what is actually happening operationally.

A standard clause copied into an agreement cannot do that work by itself.

The real risk often sits in the liability provisions

Warranties, indemnities and limitations of liability are some of the most important provisions in many commercial agreements.

They can also be among the most difficult to assess.

An agreement might contain a liability cap and therefore appear to limit risk.

But you still need to understand:

• what the cap applies to;

• whether some liabilities sit outside it;

• whether indemnities are capped;

• whether particular categories of loss are excluded;

• whether the cap is commercially appropriate; and

• whether the overall allocation of risk makes sense.

There is no universal clause that works for every business.

The appropriate position for a $20,000 consulting engagement may be very different from the position required for an enterprise software arrangement involving sensitive data and significant business-critical systems.

Why an AI-generated agreement can take longer to review than expected

There is a common assumption that if an agreement has already been drafted, a lawyer only needs to check it.

That is not always how contract review works.

If the agreement has been generated without properly understanding the transaction, the lawyer may first need to determine:

• what the commercial arrangement actually is;

• which risks need to be addressed;

• whether the existing clauses allocate those risks correctly;

• what important provisions are missing;

• whether different clauses contradict each other;

• whether the document complies with applicable laws; and

• whether significant parts of the agreement need to be rewritten.

At that point, the exercise can become less like reviewing a finished document and more like reconstructing the agreement.

In some cases, it is actually more efficient to prepare a properly tailored agreement from the beginning.

This is why an existing AI-generated draft does not necessarily reduce the amount of legal work required.

Where an agreement requires substantial restructuring or redrafting, preparing a new agreement may be more efficient than trying to fix the existing document clause by clause.

So where can AI be useful?

AI can still be an extremely useful tool.

The key is understanding what you are asking it to do.

For example, AI can be helpful for:

• explaining unfamiliar contractual terminology in plain English;

• helping you understand the general structure of an agreement;

• summarising a contract before a discussion;

• helping you organise the commercial points you want to raise;

• generating questions you may want to discuss with your lawyer; and

• helping you think through how a transaction might operate in practice.

Those are very different tasks from determining the legal position your business should adopt.

There is a distinction between using AI to help you understand a contract and relying on AI to decide how the contract should protect your business.

For an important agreement, that distinction matters.

The question is not whether AI can draft a contract

It clearly can.

The better question is whether the resulting agreement properly manages the risks associated with the transaction.

That requires understanding the business, the commercial relationship, the applicable law and the consequences if something goes wrong.

A contract is ultimately a risk-allocation document.

It determines who is responsible for what, who carries particular risks and what happens when the relationship does not go according to plan.

That is why a professional-looking document should not automatically be treated as a legally sound one.

For an important commercial agreement, the most valuable part of the process often happens before the drafting begins.

It is identifying the issues the contract actually needs to address.

When it can make sense to involve a lawyer from the beginning

If an agreement involves significant money, software, intellectual property, data, ongoing services or meaningful business risk, involving a lawyer early can often be more efficient.

A lawyer preparing the agreement from the beginning can first understand the deal and identify the areas requiring protection.

The agreement can then be drafted around those issues.

This can be much more straightforward than starting with a lengthy AI-generated document and later working backwards to determine whether the legal and commercial position is correct.

For businesses using AI extensively, the answer is therefore not necessarily to avoid AI.

It is to use it for the things it does well.

AI can help generate and explain language.

But the important legal work is often determining what the contract should say in the first place.

At Pixel Legal, we prepare commercial and technology agreements around the particular transaction, business model and risk profile rather than simply inserting standard clauses into a template.

Where an existing AI-generated agreement would require substantial restructuring or redrafting, we may recommend preparing a new agreement rather than undertaking a clause-by-clause review.

Disclaimer: This article provides general information only and does not constitute legal advice. It is not intended to be a comprehensive or exhaustive list of the legal or commercial issues that may arise when using AI to draft an agreement. The issues and appropriate contractual terms will depend on the particular circumstances. You should obtain legal advice specific to your situation before relying on or entering into an agreement.