I Let AI Review a Subcontract. Then I Reviewed It Myself.

AI is becoming part of almost every business function, and government contracting is no exception.

It can summarize information, draft correspondence, organize data, compare language, and save time on administrative tasks.

But can it actually review a government contract?

I decided to find out.

Before I Tell This Story, One Very Important Warning

I want to make this crystal clear because government contractors handle information that should never be casually uploaded into an AI tool.

DO NOT upload CUI, proprietary information, PII, export-controlled information, source-selection-sensitive information, or other protected or sensitive data into a public or unapproved AI tool.

And yes—copying and pasting information into an AI prompt counts.

If your organization uses AI with company, customer, employee, or government information, the environment should be specifically reviewed and approved by your organization's IT/security personnel for the type of information being handled.

Do not assume a tool is appropriate because:

• It's a paid subscription.

• It's marketed as an enterprise product.

• Your company already uses other products from the same provider.

• The tool says your data isn't used for training.

  
  

Know your organization's AI policy and get approval from the people responsible for your information security before putting sensitive information into any AI system.

The experiment I'm about to describe occurred in a previous position where we had access to a firewalled Microsoft AI environment that had been approved by our organization for our use.

The Experiment

We had received a new subcontract, and mostly out of curiosity, I decided to see how well our approved AI tool could analyze it.

This wasn't part of my normal contract-review process. I wanted to test the technology.

I uploaded the subcontract and asked it to look for the same types of things I would consider during a review:

• Inconsistencies

• Potential contractual risks

• Terms that appeared unusually one-sided

• Clauses or provisions that warranted additional attention

  
  

Then I waited to see what it would find.

The result?

Not much.

It gave me a handful of observations, but the items it identified weren't particularly concerning.

If someone without significant contracts experience had relied on that response, I could easily imagine the takeaway being:

Looks pretty good.

Then I performed my actual review.

The Human Review Looked Very Different

By the time I finished, the subcontract was covered in red.

There were issues throughout the document that warranted comments, questions, clarification, or negotiation.

That little experiment reinforced something I've believed throughout my career:

Reading a contract and reviewing a contract are two very different things.

The AI could analyze the document I gave it.

But a contracts professional brings something else to the table:

Context.

A Contract Doesn't Exist in a Vacuum

When an experienced contracts professional reviews a subcontract, they're not simply looking for sentences that sound unusual.

They're thinking about the entire procurement.

What did the solicitation or request for proposal say?

What did we propose?

What assumptions did we make when we priced the work?

What did we negotiate?

Does the subcontract accurately reflect those negotiations?

Did something appear in the final document that wasn't in the version we previously reviewed?

Are the prime's flowdowns actually applicable?

Are there requirements being passed down that go beyond what is necessary?

Then the review moves beyond the Contracts department.

What does this mean operationally?

Can Accounting invoice the way the subcontract requires?

Does Program Control understand the funding and cost implications?

Can the Program Manager realistically meet the reporting requirements?

Do the period of performance and deliverable dates align with the planned work?

Are there requirements that create additional cost that wasn't included in the price?

Are the termination, indemnification, intellectual property, data rights, insurance, or other business terms acceptable?

Does the document actually reflect the deal we intended to make?

Those questions require more than the ability to analyze text.

They require experience, context, judgment, and an understanding of how contractual language affects the entire business.

AI Only Knows the Context You Give It

This is especially important for small government contractors.

Imagine giving an AI tool only the subcontract and asking:

“Do you see any risks?”

It may be able to identify unusual language or summarize clauses.

But does it know that your proposal assumed three trips and the subcontract requires monthly travel?

Does it know that you negotiated a reporting requirement out and it mysteriously reappeared in the final document?

Does it know the proposed labor categories don't match the language in the subcontract?

Does it know your price was based on a particular assumption that the final agreement contradicts?

Does it know what was discussed during negotiations?

Unless you provide that context, it doesn't.

And providing all of that context potentially means feeding the AI even more company, proposal, employee, customer, or government information.

Which brings us right back to information security.

This Isn't an Argument Against AI

Far from it.

AI has tremendous potential to make contract administration more efficient.

Used appropriately—and within an organization's approved security environment—it can assist with tasks such as organizing information, summarizing lengthy non-sensitive materials, comparing versions, drafting routine language, developing checklists, and helping identify areas for further investigation.

What it shouldn't do is create a false sense of security.

“The AI didn't flag anything” is not the same thing as “this contract has been properly reviewed.”

That's the distinction small businesses need to understand.

The Temptation for Small Government Contractors

I completely understand why this technology is appealing.

A growing small business may not be ready to hire a full-time Contracts Manager.

The owner may be reviewing agreements.

The Program Manager may also be handling contract administration.

Everyone is wearing multiple hats.

Then along comes technology capable of processing a 100-page agreement in seconds.

Why wouldn't you use it?

Use the technology where it's appropriate.

Just don't confuse speed with expertise.

A polished AI response can sound incredibly authoritative while completely missing the issue that will cost your company money six months from now.

And that's not necessarily because the technology failed.

It may simply be because it didn't know what an experienced contracts professional knows to look for.

AI Should Support Professional Judgment, Not Replace It

I think that's where the real opportunity is.

The question shouldn't be:

“Can AI review our contracts so we don't need contracts expertise?”

A much better question is:

“How can technology make our contracts professionals more efficient so they can spend more time on risk, strategy, negotiation, and judgment?”

Let technology help with the administrative burden where appropriate.

Let experienced professionals decide what the contract actually means for the business.

And One More Time—Because This Matters

Do not upload sensitive information into an AI tool unless the specific environment has been approved by your organization for that information.

That includes:

CUI. Proprietary information. PII. Export-controlled information. Proposal information. Employee information. Customer information.

Don't assume.

Don't rely solely on a vendor's marketing language.

Don't decide for yourself that a system “looks secure enough.”

Ask your IT/security team. Follow your organization's policies. Know what information you're handling before you upload anything.

AI can be an incredibly useful business tool.

A preventable data-security incident is not worth finding out what it can do.

The Bottom Line

My little experiment didn't convince me that AI has no place in contract administration.

It convinced me of something much more useful:

AI and contracts expertise do different things.

Technology can process information incredibly quickly.

An experienced contracts professional understands the procurement history, recognizes risk, asks questions, challenges language, considers operational consequences, and understands what those words could mean to the business months or years after signature.

So yes, AI can read a subcontract.

But I'm still keeping the red pen.