By: Patricia McHugh Lambert, Esquire
I am, by temperament and practice, an enthusiastic fan of artificial intelligence. I use it. I recommend it. I marvel at it. But like most powerful tools—from chainsaws to email “reply all”—AI needs guardrails. Or, as I like to think of them, bumpers. Because without them, things can go sideways very quickly.
Let me begin with a confession. Recently, while reading a novel (a romance novel, because life is short and joy matters), a character died from arsenic poisoning. Another character urgently administered milk. This struck me as… questionable. So, being a modern professional, I did not head to the library or call a toxicologist. I asked ChatGPT: “Is milk an antidote to arsenic?”
What followed was not an answer, but a heartfelt intervention
The AI expressed concern for my safety, asked whether I was feeling distressed, and gently suggested resources if I was contemplating self-harm. I had to reassure an algorithm that (1) I was fine, (2) no arsenic was involved, and (3) I was simply trying to determine whether my romance novel was medically accurate. Explaining my reading habits to a machine was not on my bingo card for the day—but here we are.
Humor aside, that exchange highlights something important: AI does not understand context the way humans do, and we do not always understand how AI processes what we tell it. That gap matters—especially for business owners, and especially for regulated professionals like insurance brokers.
AI Is Public (Even When It Feels Private)
One of the biggest misconceptions about tools like ChatGPT is that they are private or confidential. They feel conversational. They feel discreet. They are not.
Public AI tools are not your company’s closed systems. Information entered into them may be logged, stored, reviewed, or used to improve the model. Even if a platform says it does not “train” on your data, that does not mean the data disappears into the ether. From a risk perspective, you should assume that anything you type into a public AI tool could someday be seen by someone else.
For insurance brokers, this is a flashing yellow light. Client information, policy details, claim scenarios, internal emails, drafting notes—none of these belong in a public AI prompt. Even anonymizing data can be risky, because context clues have a way of reassembling themselves.
AI Is Not Covered by Attorney–Client Privilege
This point cannot be overstated: asking AI for legal advice is not protected by the attorney–client privilege.
Privilege applies to communications between a client and a lawyer for the purpose of seeking legal advice. An AI system is not a lawyer. It is not your lawyer. And it certainly does not owe you a duty of confidentiality.
If a dispute ever arises—and you can assume that someday, one will—communications with AI are discoverable. That includes prompts asking, “Is this compliant?” or “How risky is this approach?” or “Can I get away with…?”
Using AI as a brainstorming tool is one thing. Using it as a substitute for legal advice is quite another—and a dangerous one.
AI Is Confident, Not Careful
Another risk is AI’s tone. AI is smooth. Polished. Articulate. And often wrong.
Large language models are designed to generate plausible-sounding responses, not guaranteed-correct ones. They do not “know” the law. They may mix outdated guidance with current rules and present it all with unwavering confidence.
Just think had AI been around in the 1600’s and AI was asked whether the world was round, its answer would have been a confident “No, the earth is flat”. Confident, but wrong.
That confidence can lull users into relying on answers they would never accept from a human without verification. And sometimes AI confidence can stifle the questioning of the status quo— and we need to sometimes question, test, explore to determine whether the accepted answer is actually correct.
The Employee Problem: When Everyone Has a Superpower
One of the quieter risks of AI is not misuse by leadership—but enthusiastic misuse by staff.
Employees may draft emails, proposals, coverage explanations, or marketing content using public AI tools without realizing the exposure. They may upload documents “just to clean up the language.”
Without clear internal guidelines, AI use becomes inconsistent, invisible, and impossible to manage after the fact.
So What Do “AI Bumpers” Look Like?
Being cautious does not mean being anti-AI. It means being thoughtful. AI is a tool, not an oracle.
Used well, AI can save time, spark ideas, and improve efficiency. Used carelessly, it can create risk that no one intended and no insurance policy wants to cover.
The goal is not to turn AI off. It is to use it with bumpers firmly in place—so it stays in the lane, and you stay out of trouble.
So here is my AI BUMPER CHECKLIST (FOR THE BUSY BUSINESS OWNER)
✓ DO assume anything entered into a public AI tool could become public.
✓ DO use AI for brainstorming, drafting, and idea generation.
✗ DON’T input client-specific, confidential, or proprietary information into public AI systems.
✗ DON’T ask AI for legal, regulatory, or compliance advice.
✓ DO require human review of all AI-generated content.
✓ DO establish clear internal rules on approved AI use.
✓ DO remember: AI is a tool, not an oracle. Judgment still matters.
And one last thing– for the record: milk is not, in fact, an antidote to arsenic. At least that is what AI told me.
Patricia McHugh Lambert is a Member in PK Law’s General Litigation Group. Ms. Lambert has over 35 years of experience in handling complex commercial litigation and insurance matters. Ms. Lambert has worked on cross-jurisdiction actions, significant litigation and regulatory matters for Fortune 500 companies. She has also assisted small and mid-sized companies and business executives with contract, real estate, liability assessment and commercial disputes that needed to be resolved quickly and efficiently. She also represents various condemning authorities in eminent domain matters, including on issues relating to scope of eminent domain powers and valuation. Ms. Lambert can be reached at 410-339-6759 or plambert@pklaw.com.
