The AI demo may be impressive.

The security review is where the harder questions begin.

Law firms are being offered more AI tools for research, drafting, summarization, document review, knowledge management, workflow automation, and client service.

Some of those tools may be useful.

But before a firm approves one, leadership should understand something the demo may not show:

What information can the AI access, what happens to that information, and what other systems or providers sit behind the product?

That question is especially timely.

The American Bar Association has an October 14, 2026 program titled How to Vet AI Tools for Your Legal Practice, focused on evaluating AI vendors, assessing security and ethics, identifying warning signs, and determining whether a tool actually fits legal workflows before adoption. Review the ABA program

The ABA describes the issue plainly: attorneys need to understand how a product secures client data, supports confidentiality, produces reliable output, and fits into real legal workflows before they invest firm-wide trust in it.

For law firms, that makes AI vendor evaluation more than a purchasing decision.

It is a confidentiality, cybersecurity, data-governance, and risk-management decision.

What Should a Law Firm Evaluate Before Approving an AI Vendor?

A law firm should understand at least five things before approving an AI platform:

  1. What data the AI can access
  2. How the vendor uses and stores that data
  3. Which third parties or model providers are involved
  4. What systems the AI can connect to
  5. What security, administrative, and governance controls are available

The exact review should reflect the tool and use case.

An AI assistant used only for non-confidential administrative work creates a different risk profile from an AI system connected to client documents, email, Microsoft 365, or a document-management platform.

That is why the question should not simply be:

"Is this AI tool secure?"

A better question is:

"Is this specific AI product, configured this way, appropriate for the information and work we intend to use it for?"

Start With What the AI Tool Can Access

Before evaluating features, start with access.

Can the AI application reach:

  • Client documents
  • Matter files
  • Email
  • Microsoft Teams
  • SharePoint
  • OneDrive
  • Document-management systems
  • Contracts
  • Litigation strategy
  • Work product
  • Internal firm documents
  • Financial or billing information

The broader the access, the more important the security and governance review becomes.

AvTek's 7 Questions Every Business Should Answer Before Expanding Its Use of AI makes the same point: organizations should understand what information they hold, who can access it, which AI tools can interact with it, and what safeguards are in place.

For law firms, those questions are particularly important because AI may interact with information subject to confidentiality obligations.

Confidentiality Should Be Part of the Vendor Review

The ABA's Formal Opinion 512 addresses lawyers' ethical obligations when using generative AI.

Among those obligations is protecting client information.

The opinion explains that lawyers using generative AI need to consider how the technology handles client information and whether appropriate safeguards are in place. Read ABA Formal Opinion 512

That means law firms should understand a vendor's actual data practices rather than assuming that an AI product marketed to lawyers automatically satisfies the firm's confidentiality requirements.

Before approval, ask:

  • What information does the vendor receive?
  • Is that information retained?
  • How long is it retained?
  • Is it used to train models?
  • Can users opt out of training?
  • Who can access submitted information?
  • What contractual protections apply?
  • What happens when an account is closed?
  • Can data be deleted?
  • How does the vendor handle security incidents?

Those answers should come from the specific vendor's current documentation and contract terms.

Ask Whether the AI Vendor Uses Another AI Provider

An AI vendor may not operate the underlying model itself.

The ABA recently highlighted that some legal AI products rely on third-party large language models such as ChatGPT, Gemini, or Claude.

That can mean information passes through more than one provider.

The ABA notes that law firms should understand those downstream data flows, particularly around model training, retention, and subsequent use.

That creates an important vendor-review question:

Who actually processes our data?

A firm may contract with one AI company while the technology relies on:

  • Another model provider
  • Cloud infrastructure
  • Additional subprocessors
  • External APIs
  • Other technology partners

That does not automatically make the tool inappropriate.

It does mean the firm should understand the chain of access.

"Enterprise" Does Not Automatically Answer Every Security Question

Enterprise versions of AI tools may provide stronger controls than consumer versions.

But the word enterprise should not replace due diligence.

The firm still needs to understand:

  • Data retention
  • Model-training terms
  • Administrative controls
  • User permissions
  • Audit logging
  • Security documentation
  • Integration scope
  • Identity controls
  • Data residency, where relevant
  • Contractual protections

A paid product may be appropriate for one legal workflow and inappropriate for another.

The security assessment should be based on the actual use case and configuration.

Review the AI Vendor's Security Documentation

Law firms should ask vendors for evidence supporting their security claims.

Depending on the vendor and service, that may include:

  • Independent security assessments
  • SOC reports
  • Security certifications
  • Penetration-testing information
  • Encryption details
  • Incident-response procedures
  • Access-control documentation
  • Privacy policies
  • Data-processing terms
  • Subprocessor lists

The ABA has noted that certifications and independent control reports do not guarantee confidentiality, but they can provide evidence that the vendor has implemented structured security and privacy controls.

That is a useful distinction.

A certification should not end the vendor review.

It can inform it.

Ask Whether the AI Tool Uses Client Data for Training

This should be a direct question.

Will our firm's data, prompts, uploaded documents, or outputs be used to train or improve an AI model?

Do not assume the answer.

Review the vendor's terms and product-specific documentation.

Law firms may also want to understand:

  • Whether training terms differ between consumer and business plans
  • Whether training is enabled by default
  • Whether an administrator can control the setting
  • Whether downstream providers have separate terms
  • Whether the policy can change over time

The ABA has specifically warned lawyers to understand how generative AI tools process and use information when client confidentiality is involved.

Review Permissions Before Connecting AI to Firm Systems

The next question is not just what the vendor does with data.

It is:

What data can the AI retrieve in the first place?

A law firm may connect AI to:

  • Microsoft 365
  • SharePoint
  • OneDrive
  • Email
  • A document-management system
  • Knowledge repositories
  • Practice-management platforms

That makes existing access controls part of the AI review.

If a user already has broad access to information they no longer need, connecting AI to that environment may make that information easier to locate and retrieve.

This is why AI governance and information governance increasingly overlap.

AvTek's existing law-firm AI guidance emphasizes the need for firms to review approved tools, client information, access rights, and review procedures before expanding AI use. Read AvTek's AI readiness guidance

Understand What the AI Can Do After It Connects

Access and authority are different.

An AI tool may be able to search or summarize information without changing anything.

Another system may be able to:

  • Create documents
  • Modify records
  • Send communications
  • Trigger workflows
  • Write information back to another application
  • Perform automated actions

Those capabilities should be evaluated separately.

Ask:

Can the AI read?

Can it write?

Can it send?

Can it modify?

Can it trigger another process?

Does a human approve the action first?

As AI tools become more connected, understanding authority becomes just as important as understanding data access.

Integration Security Deserves Its Own Review

A vendor may appear secure when viewed in isolation.

The risk can change when it is connected to the firm's systems.

For every proposed integration, ask:

  • What permissions are requested?
  • Are those permissions required?
  • Does the integration have read-only or write access?
  • Is access granted per user or across the organization?
  • Can permissions be limited?
  • How are credentials stored?
  • Can administrators revoke the connection?
  • What activity is logged?

This is especially important when the AI product connects to Microsoft 365 or other systems containing large amounts of firm and client information.

AvTek's Cybersecurity Services include identity and access controls, Multi-Factor Authentication, privileged-access management, conditional access, Microsoft 365 security, and other controls that can help firms strengthen the environment AI is connecting to.

Legal AI Should Fit the Firm's Actual Workflow

Security is not the only part of vendor evaluation.

The ABA's October 14 program also emphasizes determining whether a tool genuinely fits the firm's workflows rather than simply performing well in a polished demonstration.

That is important because a tool that employees do not use consistently can create its own problems.

The firm may end up with:

  • Multiple overlapping AI tools
  • Employees returning to unapproved applications
  • Data scattered across platforms
  • Inconsistent workflows
  • Additional vendor relationships
  • More systems for IT to manage

A useful AI tool should fit the way the firm actually works.

That can include compatibility with:

  • Existing document-management systems
  • Microsoft 365
  • Practice-management software
  • Matter workflows
  • Existing security controls
  • The firm's review process

AvTek's existing article on AI risks for law firms discusses why uncontrolled AI adoption can create problems involving confidentiality, cybersecurity, inconsistent usage, and disconnected workflows.

Do Not Let a Successful Pilot Bypass Vendor Governance

AI tools often enter organizations through experimentation.

One attorney tries a product.

Then a few more employees use it.

The firm finds a useful workflow.

Soon the product is being used with real matter information.

That is a point where an informal experiment can become a production technology decision.

The firm should know when that transition requires:

  • Security review
  • Vendor review
  • Contract review
  • Data-access review
  • Leadership approval
  • IT involvement
  • AI-governance approval

A pilot that uses non-confidential information may not require the same controls as a deployment connected to client systems.

The review should evolve with the use case.

AI Vendors Should Be Part of the Firm's Third-Party Risk Conversation

AI providers are still vendors.

If a vendor receives confidential data, connects to firm systems, or becomes important to daily operations, it should be considered within the firm's broader third-party risk process.

Questions may include:

  • What happens if the vendor experiences an outage?
  • How does the vendor notify customers of a security incident?
  • Can the firm retrieve its data?
  • What happens if the provider changes ownership?
  • What happens if the vendor changes model providers?
  • Does the vendor materially change its terms?
  • How quickly can the firm disable access?
  • What happens when the relationship ends?

For law firms that need additional cybersecurity leadership around technology and vendor decisions, AvTek's vCISO Services include cybersecurity risk assessments, governance, policy development, vendor and third-party risk discussions, security recommendations, and executive guidance.

AI Governance Creates a Repeatable Approval Process

Individual vendor reviews are useful.

A repeatable approval process is better.

AvTek's AI Governance Services are designed to help organizations establish:

  • AI risk assessments
  • Approved AI tools
  • Acceptable-use standards
  • Data-protection guidance
  • Roles and responsibilities
  • Processes for approving new AI applications
  • Ongoing governance as technology changes

That gives a firm a consistent answer to:

Who decides whether we can use this AI tool?

The objective is not to block every new application.

It is to avoid making the same governance decision from scratch every time someone finds a new AI product.

What Should Be on a Law Firm AI Vendor Security Checklist?

Before approving an AI vendor, a law firm should consider documenting answers to the following questions.

1. What Is the Business Use Case?

What problem is the AI tool intended to solve?

2. What Data Will It Access?

Will it process confidential client information, matter documents, work product, or internal firm data?

3. Is Firm Data Used for Model Training?

What do the vendor's current terms say?

4. How Long Is Data Retained?

Can the firm control or delete retained information?

5. Which Other Providers Are Involved?

Does the vendor use third-party models, cloud providers, or subprocessors?

6. What Systems Can It Connect To?

Will it have access to Microsoft 365, email, SharePoint, OneDrive, a DMS, or another system?

7. What Permissions Does It Receive?

Is access limited to what the product actually needs?

8. Can the AI Take Actions?

Can it only read information, or can it modify, send, create, or trigger activity?

9. What Administrative Controls Are Available?

Can the firm control users, permissions, integrations, retention, and approved features?

10. What Logging Is Available?

Can administrators review relevant AI activity?

11. What Security Evidence Does the Vendor Provide?

Are independent assessments, certifications, or control reports available?

12. How Does the Vendor Handle Security Incidents?

What notification and response procedures are documented?

13. How Does the Product Fit Existing Firm Workflows?

Will employees realistically use the approved platform?

14. Who Owns the Vendor Relationship?

Who monitors changes to the product, contract, and data practices?

15. What Happens When the Firm Stops Using It?

How are data, accounts, integrations, and permissions removed?

Those questions turn vendor selection from:

"Do we like the product?"

into:

"Can we responsibly use the product for this specific purpose?"

What Are Red Flags When Evaluating a Legal AI Vendor?

There is no single red flag that automatically disqualifies every technology vendor.

But firms should pay closer attention when important questions produce unclear answers.

Examples may include:

  • Vague descriptions of data use
  • Unclear model-training terms
  • No identifiable security documentation
  • Unclear subprocessors
  • Very broad integration permissions
  • Limited administrative controls
  • No clear incident-notification process
  • No ability to control or remove firm data
  • Marketing claims that cannot be explained technically
  • A product that requires far more access than the use case appears to need

The ABA's current AI-vendor programming specifically emphasizes recognizing warning signs early rather than discovering them after firm-wide adoption.

The Vendor Review Should Continue After Purchase

AI products change quickly.

A vendor may:

  • Add new features
  • Change an underlying model
  • Add new integrations
  • Introduce AI agents
  • Change retention practices
  • Add subprocessors
  • Modify contract terms
  • Expand the data available to the product

That means vendor approval should not necessarily be treated as permanent.

The firm should determine which changes trigger another review.

AvTek's AI governance model specifically includes ongoing review as AI use, business needs, risks, and technologies change.

Frequently Asked Questions About AI Vendor Evaluation for Law Firms

What should a law firm ask an AI vendor before buying?

At minimum, ask what data the tool accesses, whether firm data is retained or used for training, which third parties process the data, what systems the product connects to, what permissions it receives, what administrative controls are available, and how security incidents are handled.


Can a law firm put confidential client information into an AI tool?

The answer depends on the specific product, configuration, data-handling terms, client information involved, applicable ethical obligations, and safeguards in place.

ABA Formal Opinion 512 advises lawyers to consider confidentiality obligations and understand how generative AI tools handle information before using them with client data. Read ABA Formal Opinion 512


Should law firms ask whether an AI vendor trains on their data?

Yes.

Firms should understand whether prompts, documents, outputs, or other submitted information may be used to train or improve AI models and whether those practices differ between product tiers.


What is a subprocessor in an AI product?

A subprocessor is another provider involved in processing data on behalf of the primary vendor.

For AI services, this may include cloud providers, model providers, or other technology companies involved in delivering the service.


Should law firms review AI integrations separately?

Yes.

Connecting an AI tool to Microsoft 365, a document-management platform, email, or another business system can change what information the product can reach and what actions it can perform.

Permissions should be reviewed as part of the integration.


Does SOC 2 mean an AI vendor is safe to use?

A SOC report can provide evidence about a vendor's control environment, but it does not by itself determine whether a specific AI product is appropriate for confidential legal work.

The firm should still review data practices, permissions, integrations, contracts, and the intended use case.


Who should approve AI vendors at a law firm?

The appropriate approval process depends on the firm.

Depending on the product and use case, review may involve IT, cybersecurity, firm leadership, legal operations, risk management, privacy, or other responsible parties.

The important thing is that the firm has a defined process and clear ownership.


How often should law firms reevaluate AI vendors?

There is no universal schedule that fits every vendor.

Firms should consider reassessment when material changes occur, such as new integrations, model changes, new data access, altered contract terms, new subprocessors, or expanded use across the firm.

The AI Demo Shows What the Product Can Do. The Vendor Review Shows What It Can Touch.

Law firms should not evaluate AI solely on productivity.

A useful vendor review should answer:

What information can the tool access?

How does the vendor use that information?

Who else processes it?

What systems can the AI connect to?

What permissions does it receive?

Can it take actions?

Can administrators see what it is doing?

What happens when the vendor changes the product?

Those questions are increasingly important as AI moves from standalone chatbots into connected legal workflows.

The goal is not to prevent law firms from using AI.

It is to make sure the firm's governance keeps pace with what the technology is being allowed to do.

AvTek Solutions helps law firms connect AI adoption with cybersecurity and risk management through AI Governance Services, Cybersecurity Services, Co-Managed IT Services, and vCISO Services.

If your firm is evaluating an AI product, the best time to ask about data access, security, permissions, and vendor risk is before the tool becomes part of everyday legal work.

Talk with AvTek about AI governance and cybersecurity for your law firm