What AI Changed About the Security Questions Worth Asking
Almost every firm I talk to now has people usingAI tools during the working day. Drafting, summarizing, cleaning up a spreadsheet, turning a rough set of notes into something presentable. Very few of those firms decided to allow it. It simply happened, the waycloud storage happened about a decade earlier.
I am not going to tell you to ban it. That advice does not survive contact with a busy team, and it was not good advice anyway. The useful questions are narrower and more answerable than the conversation usually suggests.
Four questions that are worth an hour of your time
Which tools are in use, and who is paying for them
Start with the inventory, because most firms do not have one.Personal accounts are the common case, and a personal account means the firm has no visibility, no record, and no way to remove access when somebody leaves. That is an ordinary access control problem rather than an AI problem, and it has ordinary fixes.
What happens to what gets pasted in
Consumer versions of these tools and business versions of the same tools handle submitted content differently. That distinction is the single most consequential thing to understand, and it is knowable. A firm holdingclient financial records, patient information, or customer data held under contract should know the answer for every tool in use before deciding what is acceptable to put into it.
What your own client agreements already say
Before worrying about regulation in general, read theconfidentiality language you have already signed. Many professional services agreements contain commitments about subcontractors andthird party processing that were written years before any of this existed but apply to it cleanly. That is usually where the real constraint sits.
Is there an AI acceptable use policy that everyone has bought into
Most firms do not need a long document. They need an AIacceptable use policy that settles three things: which tools are approved, what data may and may not go into them, and the security requirements every tool has to meet, such as business accounts and multifactor authentication. It should also say who to ask when the answer is unclear. The part that matters most is buy in. A short policy that everyone has read and agreed to is worth considerably more than a long one that sits in a shared drive.
What I would put in place first
If a firm asked me to spend one day on this, I would not start with anything branded as AI. I would start with the controls that make any of it manageable.
Multifactor authentication on every account that touches firm data, without exceptions for anybody senior.
Business accounts rather than personal accounts for anything used for work, so access can be granted and removed centrally.
A current list of who has administrator rights, and a short list of people who should.
Endpoint detection running on the machines people actually work on, including home machines if those are in use.
The AI acceptable use policy described above, approved by leadership and acknowledged by everybody who uses these tools.
That is not a sophisticated program. It is the foundation that everything more sophisticated depends on, and most firms are further from it than they expect.
On the numbers you are going to see this month
October brings a reliable wave of alarming statistics about small business and cyber risk. Some are sound. Many trace back to a source that does not say what the headline says it says. I do not cite figures I have not checked, and I would encourage you to ask anybody who does cite one where it came from.
The case for doing this work does not require a statistic. It requires an honest look at what your firm holds and who currently has access to it.
There is a session on this next week
I am hosting a local session on this on Wednesday, October 14, at the Commander’s Mansion in Watertown. The guest speaker is Matt Topper, Evangelism Director at ConnectWise, who holds CISSP, CISM, and CCSP and leads IT Nation Secure. The format is one topic covered properly rather than a tour of everything, and the goal is that people leave with something they can act on that week. It runs from 4:30 pm to 6:30 pm, it is free, and registration is required. Details and registration are below.
If you would rather just talk it through, call me at (617) 221-4100.
Frequently Asked Questions
1. Should my firm ban AI tools? +
Usually not. A complete ban can be difficult to enforce and may push employees toward personal accounts or unapproved tools that the firm cannot monitor. A more practical approach is to provide approved AI tools, clear usage rules, and defined requirements for handling business and client data.
2. How do I find out which AI tools my employees are already using? +
Start with an inventory. Ask teams which tools they use, review browser extensions, software subscriptions, cloud application permissions, and expense reports, then maintain an approved tools list so the information stays current.
3. Why are personal AI accounts a problem for businesses? +
Personal accounts can limit the firm’s visibility into what information was submitted, how the tool was used, and whether access can be removed when an employee leaves. Business-managed accounts provide stronger administrative control and accountability.
4. What is the difference between consumer and business versions of AI tools? +
Consumer and business plans may differ in how submitted information is stored, retained, accessed, logged, or used to improve services. Businesses should review the terms and security controls for each plan before deciding what information employees are allowed to enter.
5. Do small firms really need to worry about AI risk? +
Yes. AI-related risk comes from how tools are used and what information employees enter, not just from company size. Any firm handling client, customer, financial, legal, healthcare, or other sensitive information should manage AI use carefully.
6. What types of data should never be pasted into an unapproved AI tool? +
Businesses should be especially cautious with client financial records, patient information, confidential customer data, credentials, proprietary information, and any content protected by contracts, privacy rules, or internal policy. The exact rule should be defined for each approved tool and data type.
7. Where should I look first for AI-related restrictions? +
Start with existing client agreements, confidentiality provisions, vendor requirements, privacy obligations, and internal policies. Contract language written before modern AI tools existed may still restrict how client or customer information can be shared with third parties.
8. Can using an AI tool count as sharing data with a third party? +
It can, depending on the tool, contract terms, data involved, and how the service processes information. A third-party risk review can help determine whether an AI platform is appropriate before employees begin using it with business data.
9. Does AI use affect state data protection requirements? +
It can. Businesses should consider how AI tools fit within applicable state privacy and data protection requirements. Massachusetts firms, for example, should review how AI use aligns with their obligations for protecting personal information.
10. How can I prove to clients that we use AI responsibly? +
Maintain clear documentation such as an approved AI tool list, acceptable use policy, employee acknowledgments, account settings, vendor reviews, and security controls. Having this information available can make client security reviews and questionnaires much easier to complete.
11. What should an AI acceptable use policy include? +
At minimum, the policy should identify approved AI tools, explain what data employees may and may not enter, define basic security requirements, and name the person or team employees should contact when they are unsure whether a use case is allowed.
12. How long should an AI policy be? +
It should be long enough to establish clear rules but short enough that employees will actually read and understand it. A concise policy that employees acknowledge and follow is often more useful than a lengthy document that is rarely referenced.
13. Who should approve the AI policy? +
Leadership should approve the policy so employees understand that responsible AI use is a business requirement rather than an optional IT preference. Depending on the organization, legal, compliance, HR, or security teams may also contribute.
14. How do I get employees to follow an AI policy? +
Explain why the rules exist, provide approved tools that are easy to use, train employees on acceptable data handling, and collect written acknowledgments. Practical guidance and good alternatives usually drive better adoption than restrictions alone.
15. What security controls should come before AI-specific measures? +
Strong foundational controls should come first, including multi-factor authentication, business-managed accounts, limited administrator access, endpoint protection, patching, monitoring, and a clearly documented acceptable use policy.
16. Why does multi-factor authentication matter for AI security? +
Multi-factor authentication helps protect the accounts employees use to access business data and approved AI tools. It is most effective when it is applied consistently across users, administrators, and senior leadership without unnecessary exceptions.
17. Why should businesses limit administrator rights? +
Reducing the number of administrator accounts limits the potential impact of a compromised account or accidental change. Businesses should regularly review administrator permissions and remove elevated access that is no longer necessary.
18. Does endpoint protection need to cover home computers? +
Yes, if home computers are authorized to access firm data or business systems. Any device used for work should meet the organization’s security requirements, including endpoint protection, patching, access controls, and monitoring where appropriate.
19. How long does it take to get started with AI governance? +
The basic steps can begin quickly. Businesses can start by enabling multi-factor authentication, moving work to managed business accounts, reviewing administrator privileges, identifying current AI tools, and drafting a practical acceptable use policy.
20. Who is the October 14 session for, and does it cost anything? +
The October 14 session is designed for owners and managers of local firms who want practical steps they can apply immediately. It is scheduled from 4:30 PM to 6:30 PM at the Commander’s Mansion in Watertown. Attendance is free, but registration is required.
