Your Chatbot Can’t Keep a Secret: Why Employers Still Need Lawyers (For Now)

Last week, we held our annual labor & employment seminar before a full house (!) at Hotel Marcel in New Haven. In my intro, I made an off-the-cuff joke about how we were thankful for those in attendance as we know that they had their AI Chatbots they could always ask.

It got laughs, but as I was thinking about this (and how I will never be a stand up comic), it got me thinking: When DO employers need attorneys anymore and why?

I’m not the first person to ask the question, and I won’t be the last. I’ve been poking at it on this blog for a while. Back in 2022, I asked an early version of ChatGPT to explain Title VII to a fifth grader and to tell me whether an employer could withhold pay from an employee who stole food. The answers weren’t bad. My conclusion at the time was that AI may not replace attorneys yet, but its ability to provide basic information was clearly improving.

Four years later, “clearly improving” undersells it. So let’s try to answer the question honestly, without the reflexive defensiveness you might expect from a lawyer whose livelihood depends on a particular answer.

First, Some Credit Where It’s Due

Let’s start with what these tools do well, because they do quite a lot well.

Adoption isn’t a fringe phenomenon anymore. A July 2026 survey by the Association of Corporate Counsel found that 85% of in-house legal departments now deploy AI in their practice, up from 53% a year earlier. Thomson Reuters reports that 74% of professionals now use AI tools several times a week. On the HR side, SHRM found that 39% of organizations have AI adopted in their HR functions, and 87% of HR professionals using it report improved efficiency.

What are people using it for? Mostly the right things. Summarization tops the list for corporate legal departments at 83%, followed by contract clause identification and first-pass review. In-house lawyers cite efficiency gains in drafting and legal research as the biggest benefits.

That tracks with my own experience. A good AI tool, like the one we use (Harvey) can take a 150-page deposition transcript and give you a usable summary in minutes. It can produce a decent first draft of a policy, an outline for a training, or a list of issues to think about before a hard conversation with an employee. It can explain the difference between FMLA and CT Paid Leave in plain English. If you’re an HR professional who used to call outside counsel to ask “what does this term mean,” you probably don’t need to anymore. Frankly, that’s fine with me.

So if the question is whether AI can handle a lot of the routine informational work that used to land on a lawyer’s desk, the answer is yes. The harder question is what happens at the edges.

The Law Moves Faster Than the Model

Here in Connecticut, October 1 was a busy day. Public Act 26-12, the omnibus labor bill signed by the Governor on May 11, 2026, took effect for most purposes. Among other things, all employers must now include the wage range and a general description of benefits in internal and public job postings. The ban on employment promissory notes now applies to every employer, not just larger ones. Employers have a new obligation to give written notice of the right to reasonable accommodation, including to existing employees within 120 days. Lactation breaks must now be provided in addition to regularly scheduled breaks.

Then there’s the state’s new AI law, Public Act 26-15. Effective October 1, 2026, an employer may not use reliance on AI as a defense to a claim under the Connecticut Fair Employment Practices Act, and the CHRO and courts may consider anti-bias testing or similar proactive efforts when analyzing claims involving automated decision tools. Employers serving WARN notices must also disclose whether AI was involved in the layoff decision, and the broader notification requirements kick in on October 1, 2027. Add in revisions to the electronic monitoring law, a new rule making employer-required security screening time compensable, and a minimum wage increase to $17.48 on January 1, 2027.

Now ask yourself: what does a general-purpose chatbot know about any of that? Depending on when its training data ends and whether it searches the web, it may well describe Connecticut’s 2021 pay transparency rule, which only required disclosure upon request or at the time of an offer. It will sound confident either way.

And that’s just Connecticut. If you have employees in Massachusetts, New York, and Rhode Island, each has its own rules. Jurisdictions across the country are separately regulating automated employment decision tools, which I’ve called a compliance nightmare for multi-state employers without a federal law to preempt them. Knowing which rule applies to which employee, in which location, as of which date, is where a lot of the real legal work sits.

Confident and Wrong

You’ve probably heard about lawyers getting sanctioned for citing fake cases generated by AI. The problem hasn’t slowed down. It’s tempting to think the specialized legal research tools solved this. They haven’t, at least not entirely.

In many instances, the lawyers who got burned did not do the job a lawyer is supposed to do, which is check the work. If an HR manager relies on a chatbot’s description of a statute to make a termination decision, there’s no one in the loop to check it at all.

Your Chatbot Can’t Keep a Secret

This, to me, is the most underappreciated reason employers still need lawyers.

And courts are still grappling with this question: Are a person’s exchanges with a generative AI platform protected by the attorney-client privilege or the work product doctrine? Several courts have held that AI isn’t an attorney, that the platforms’ privacy policies undercut any expectation of confidentiality, and that the work wasn’t done at counsel’s direction.

The law here is still developing. But for an employer, the practical lesson is the same. If your HR director types the facts of a harassment complaint into a public chatbot and asks “can we fire this guy,” you should assume that conversation is discoverable.

I’ve written about how that cuts both ways. Employees are using these tools to draft complaints and demand letters. Their chat histories are fair game in discovery and may contain valuable impeachment material. Employers, in turn, need to update litigation holds to cover ChatGPT, Claude, and Copilot data.

A conversation with your employment lawyer, on the other hand, is privileged. That protection is one of the oldest in the law, and it attaches to a human relationship that no software license creates.

Agencies, Investigations, and Judgment Calls

Beyond research and privilege, there’s a category of work that is hard to hand off to a machine because it isn’t really about information at all.

Take a CHRO complaint. The legal standards are knowable. But deciding what to say in a position statement, what documents to attach, how to handle a fact-finding conference, and when to consider settlement are strategic calls (at least for now). Whatever you say to the agency will follow you into court if the case goes there. The same is true for a DOL wage audit or an NLRB charge, where the governing standards can swing significantly from one Board to the next.

Internal investigations are similar. Interviewing a frightened complainant, assessing a witness’s credibility, deciding whether to structure the investigation to preserve privilege, and writing findings that will hold up in litigation are deeply human tasks. It’s telling that general counse, remain uncomfortable using it for investigations and privilege review.

And then there are the judgment calls. Should you terminate the long-tenured employee who just requested FMLA leave and also has documented performance problems? How do you select employees for a reduction in force without creating a disparate impact problem? What’s the reputational risk if this separation becomes a story? An AI tool can list the factors. It can’t sit across the table from your CEO and tell them, with conviction and some accountability, that this is the wrong week to do this.

Knowing the law, it turns out, is only part of the job. Temperament, perspective, and knowing when to push and when to let something go are what clients are really paying for.

So When Should You Call?

Here’s my honest take. Use AI for the first draft, the summary, the background explainer, and the issue-spotting list. You’ll save time, and your lawyer bills will probably go down. But call your employment counsel when:

• Something has been filed or threatened, whether with a court, the CHRO, the EEOC, the NLRB, or the DOL.
• You’re about to make a decision with real legal or reputational stakes, like a termination involving protected activity, a RIF, or an executive separation.
• You’re conducting or planning an investigation.
• The question involves a new or changing law, especially one that took effect in the last year.
• You’d be uncomfortable seeing your question, and the answer, as an exhibit in a lawsuit.

That last one is the simplest test I can offer.

So back to the joke. Yes, everyone in that ballroom at Hotel Marcel had a chatbot in their pocket. They came anyway. I suspect it was because they wanted to hear from people who had lived through the cases, sat across from the agency investigators, and read the new statutes before they took effect.

Your chatbot can tell you what the law says, most of the time. When it matters, you want someone who will tell you what to do, keep the conversation privileged, and stand behind the advice.

And if you missed this year’s seminar, there’s always next year. We’ll save you a seat. Bring your chatbot too.