Skynet wasn’t such science fiction after all

How many of you have encountered AI notetakers out in the wild yet?

Even if you’re not using them at your own workplace, chances are you’ve been recorded by one by now. They’re extremely popular with doctor’s offices in particular, where your consent for such recording can be subtly obtained alongside the customary “unwieldy pile of intake forms,” slipped somewhere behind the boilerplate HIPAA acknowledgement you always pretend to skim, just to show the receptionist that you’re a very serious person. 

To be fair, notetaking might be one of the ways in which AI has a practical application… at least in theory. Frankly, the bar for AI is pretty low at this point in time, because – and if you take just one thing away from this article, please let it be this – so-called “AI” is literally just glorified autocorrect, and attempting to use it for any serious purpose, such as business emails or (heaven forbid) actual contracts or other employment documents, will not come to a good end.

That AI-drafted contract might look good at a casual glance, but you know who doesn’t take “casual glances” at contracts? Plaintiff’s attorneys, judges, DOL auditors, EEOC investigators, etc.; in the event of a dispute, no amount of pretty formatting or nonsensical legalese will be able to hide the fact that your “contract” was written by the same program that your phone uses to insist that you surely must have meant to type “ducking” for the thousandth time. 

So, if AI can’t draft a decent legal document – or even a casual email, for that matter – what exactly is it good for? At the risk of editorializing, creating videos of animals jumping on trampolines, political cartoons, and deepfakes seem to rule the day.  But our entire economy is betting on it, so any vaguely useful application will be touted as the next evolution of the iPhone. Which brings us to the topic at hand: AI notetaking, and the class action lawsuit that might just bankrupt multiple businesses that just had to jump on the AI bandwagon without policies, training, and proper planning. 

AI notetakers are purportedly exactly that: they’re a computer program that records and summarizes a meeting. They’re convenient, and fairly accurate (as AI programs go), so why all the snarky leadup? Why clarify “purportedly” at the start of this paragraph? Well, because while AI notetakers do indeed take notes, that’s not what they’re actual purpose is. Their actual purpose is surveillance and “training” data for the AI company. Because no matter how many forms you sign – and even if you have the premium pad version of one of these HAL 9000 monstrosities – not a single word spoken into a Chat GPT, Claude, or Grok AI notetaker is private as these are open systems. 

Client names, legal strategies, medical information, social security numbers, you name it: each and all are recorded straight onto an open AI notetaker system such as Chat GPT, Claude, or Grok AI are then kept on that company’s servers and used to “train” the very AI program you’re using as a transcriber. Ah well, what’s the big deal? These are multi-billion dollar computer companies, surely their encryption is best-in-class?

It’s not. It’s actually remarkably vulnerable to, and routinely infiltrated by, hackers. It’s also routinely sold, for a profit, to anyone willing to pay, including advertisers and yes, even the US government. But that’s all to be expected. As everything entered into Notes becomes a part of the AI’s language model, your data will be used in other customers’ AI-generated documents. As in, when the AI needs to invent a placeholder name or social security number for an unrelated user, it might “decide” to use the data it recorded from you, thereby implicating you in a data breach that could end in class action lawsuits, bankruptcy, and even criminal charges.

Legal scholars are advising that any of the following subjects could open the floodgates of liability for AI-users: consent; biometrics; discrimination and disparate impact; attorney-client privilege; data retention; and confidentiality. 

There’s much to unpack here, but by way of example, many huge multi-national companies have begun using AI to screen job applications. Seems reasonable, right? When you’re getting 100’s of applications any given day, anything that might help you sift through them would be an asset. The problem is that these AI programs have been caught doing things like automatically rejecting applications from specific protected categories. Now, if a human being systemically shredded all applications from black applicants, that’s obviously legally actionable discrimination. But who is liable when an AI program does the same thing? Because we all know it won’t be the AI companies themselves, so who does that leave…?

Don’t be caught holding the proverbial bag when this all comes crashing down.  AI is not going away; we need to learn how best to govern and manage it, just never ever lose sight of the fact that nothing entered into an open AI program is secure in any sense of the word.  Inputting certain data into these open programs might be unlawful in and of itself, even before we get to issues like data brokers selling you and your clients’ information to any willing buyer. 

As was said by Arnold in Terminator 2: Judgement Day': 

John Connor: "We're not gonna make it are we? People I mean." 

The Terminator: "It's in your nature to destroy yourselves." 

In the meantime, we’ll be keeping a close eye on In re Otter.AI Privacy Litigation as it works its way through the courts and updating you as concrete legal standards and conclusions are set forth, motion-by-motion, hearing-by-hearing. Stay tuned!