The wearable wiretap.

Dave Bittner: Hello, everyone. And welcome to "Caveat," N2K CyberWire's privacy, surveillance, law, and policy podcast. I'm Dave Bittner and joining me is my co-host Ben Yelin from the University of Maryland Center for Cyber Health and Hazard Strategies. Hey there, Ben.

Ben Yelin: Hello, Dave.

Dave Bittner: On today's show Ben discusses potential conflict between the new generation of wearable technology and state wire tapping laws. I've got the story of a legal victory for AI coders. And later in the show my conversation with Danny Manimbo, ISO and AI practice leader at Schellman. We're discussing why companies waiting for AI regulatory clarity are already behind. While this show covers legal topics and Ben is a lawyer, the views expressed do not constitute legal advice. For official legal advice on any of the topics we cover please contact your attorney. All right, Ben. It's great to see you, but before we jump in to today's stories I need a bit of clarification from you.

Ben Yelin: Sure thing.

Dave Bittner: I was going through my social media, happened to be over on Mastodon which is where I like to hang out, and saw a posting from your future boyfriend Orin Kerr who posted the Chatrie geofencing case is now on remand back to the fourth circuit for argument on the legality of the warrant.

Ben Yelin: Sounds like a different language to you. Huh?

Dave Bittner: Can you please -- [laughs] I was going to say can you please translate this one. What the heck does this mean?

Ben Yelin: Sure. So the Chatrie case went to the Supreme Court and it was decided on the narrow question of whether there was a search in the first place.

Dave Bittner: Okay.

Ben Yelin: So what the Supreme Court decided is that there is a search when you're talking about the type of location collection that was taking place here for the geofence warrants.

Dave Bittner: Right.

Ben Yelin: That was the only issue that the Supreme Court resolved. There's the secondary question of, well, if there is a search, was there a proper warrant issued based on probable cause?

Dave Bittner: Oh.

Ben Yelin: And that question was left unresolved by the Supreme Court. They basically said "We did our part in determining whether this was a search in the first place. It is. So in order to determine the legality of the warrant we are remanding this to the lower court and they will hear the merits of the argument from both sides on the legality of the warrant. So.

Dave Bittner: What would be the outcome of that?

Ben Yelin: The outcome of that would be either the warrant itself was defective in which case Mr. Chatrie is probably still going down no matter what.

Dave Bittner: Okay.

Ben Yelin: Because if the warrant was defective they'll just say they were acting in good faith and I think the good faith exception would end up applying. Or they'll say that the warrant which ended up being issued, this geofence warrant, is not defective. There's a lot of disagreement among circuits about whether you can even have a proper warrant in a geofencing case. The fifth circuit said you can't, that warrant requires a particularity and by definition when you're doing a geofence warrant you're asking for a bunch of different devices so you do not have particularity. But this is something that the fourth circuit is going to have to decide for itself in this case.

Dave Bittner: Is this like a -- basically the Supreme Court throwing a boomerang at the fourth circuit knowing that it will come back to them?

Ben Yelin: It might come back to them. It might not. I think it's certainly possible, especially if they disagree with the fifth circuit or if other circuits chime in and there's a broad disagreement on whether this type of warrant can be issued. But the Supreme Court only likes to resolve issues if they have to resolve an issue and they are perfectly content to just punt something down to the lower court to avoid making a decision when they don't have to.

Dave Bittner: Okay. Fair enough. Those lazy Supremes. Right?

Ben Yelin: I mean I hate to use the lazy word, but they do only take like, you know, 80 cases a year. Compared to some of these circuit courts who are taking, you know, multitudes of that, so --

Dave Bittner: Well, you know [laughs]. All right. Well, thank you for the clarification there. I honestly could not make heads or tails of what it actually meant. So I appreciate you unpacking it for us.

Ben Yelin: Absolutely. Any time.

Dave Bittner: Ben, first of all, before we dig in to our stories here, cause for celebration. This is the first show of season seven of "Caveat."

Ben Yelin: Can you believe it?

Dave Bittner: Seven seasons. Lucky seven.

Ben Yelin: Lucky number seven. I didn't think they would let us go one full season.

Dave Bittner: I know. I know. And here we are.

Ben Yelin: It's been an honor. Happy to be back for season seven. In the sitcom world this is usually when the show starts to go downhill, you know.

Dave Bittner: That's true. We're going to jump the shark.

Ben Yelin: It's they have to start getting gimmicky or, you know, the star leaves.

Dave Bittner: We need to bring in a cute kid to refresh things and --

Ben Yelin: Oh. That's a good idea.

Dave Bittner: Like Oliver on the "Brady Bunch."

Ben Yelin: Oliver on the "Brady Bunch." Cousin Oliver. Yeah. Ethan might be our cousin Oliver.

Dave Bittner: Oh. There you go. Ethan is our cousin Oliver.

Ben Yelin: He gets a -- he gets a shout out. And that's intended to be an honor, Ethan.

Dave Bittner: Okay. Well, thanks to all of our listeners for making this possible. If it weren't for all of you, we certainly would not be doing this every week. So we do appreciate you taking the time to spend the time with us. We do genuinely appreciate it. All right. Well, let's jump in to some stories here, Ben. Why don't you kick things off for us?

Ben Yelin: So I was at the mall last week with my kids and walked by a store that was advertising the new Meta smart glasses.

Dave Bittner: Oh yes. I -- my most recent eye doctor appointment my eye doctor had them as, you know, frames you could choose.

Ben Yelin: I mean it's a thing now. My kids were fascinated by it. I didn't have much information to tell them except that, you know, it strikes me as a little bit creepy having these AI powered glasses. But these wearables are advancing at lightening speed. All of the companies are producing their own wearable. So you have Amazon's Bee Frontier and Bee Eclipse devices, Apple's new Apple Watch feature such as Siri recap and live rewind. We'll talk about the implications of that. And then you have other startups. So wearables have evolved from what they used to be. We're long past the days of Google glass, if you can recall that.

Dave Bittner: The glass holes.

Ben Yelin: So there is a potential conflict here when we're talking about two party consent state laws.

Dave Bittner: Right.

Ben Yelin: So some states have one party consent where you can record surreptitiously. I botched that pronunciation. But you can record somebody if just one of the parties has consented to it. Maryland is a two party consent state as are many states across the country. So in order to secretly record somebody both parties have to consent to it. Otherwise it violates Maryland's wire tapping law.

Dave Bittner: Right.

Ben Yelin: These devices present a novel problem. So most old wearables and other voice assistants like your standard Alexa rely on a wake word. So they are asleep. They are snoozing. They are risen as soon as you say the wake word. The wake word might trigger them to record, you know, the two or three seconds before the wake word and then going forward through the recording. But these wearables are designed to listen continuously and to extract meaning from conversations around them even if the audio is not saved. So if the device is consistently listening and generating summaries without ever saving the actual recordings there's this novel legal question as to whether it's been intercepted as the law would define it a private conversation under state wire tap laws. So CNBC had an article on this. They interviewed a bunch of attorneys, legal scholars. State law seemed to focus more on whether a communication was acquired or intercepted. So that -- those are very specific terms of art. You know, it's somebody actually taking possession of a communication. Seemingly this was designed for the era of actual tape recording. So you get the tape or you steal the tape.

Dave Bittner: Right.

Ben Yelin: Now you have something different where in some cases an audio file is stored. In other cases it's converting speech that it hears in to not speech to text, but speech in to text summaries. And that presents just a really interesting legal concern that our courts are just not set up to deal with. Now the legal scholars who were interviewed for this I think correctly said like, "It's very unlikely we're ever going to get a case that comes out of this." It would have to be an extremely unlikely scenario where somebody says something incriminating while a wearable is listening to them. The wearable generates a summary. The user seemingly sees that summary and alerts law enforcement. Then that might be the fruits of the poisonous tree. Again that's a pretty unlikely scenario, but eventually that type of thing is going to happen. So the concern here is that people could find themselves being monitored without ever having consented to being monitored. Obviously the person who has the wearable has consented to being monitored. You know, they are turning on features. They're signing a EULA. But the bystander has not agreed to it and sometimes it might be the bystander who suffers the consequences. So it's just a really interesting novel issue. For their part, Apple and Amazon say their systems are designed to protect privacy. They do not store raw audio. And they limit access to user data. But these aren't foolproof safeguards just because you do have this capability to transcribe summaries of audio that's being heard from these wearables. This is one of those, you know, the technology's moving much faster than privacy law and states might have to end up adjusting two party consent laws to account for this scenario if it becomes a bigger problem.

Dave Bittner: How does two party consent work with things like security cameras? Everybody has a nest camera on their front door these days and, you know, there are cameras all over the place. Does that generally fall in to the no expectation of privacy in a public place thing?

Ben Yelin: Yes. It does. So just to, you know, add some color to that, we had an incident here in Maryland where a member of Congress was having a town hall and after the town hall he held a press conference.

Dave Bittner: Oh yeah.

Ben Yelin: And there were like 10 members of the press there. One of them was a University of Maryland journalism student who I think had asked a really reasonable question. And the congressman got very angry at her and said, "You're violating the law. This is a two party consent state. Turn off your recording device now." And she's like, "But all these other people are recording you." And he's like, "I don't care. Two party consent. Turn it off." Eventually he like had his body guard go get law enforcement to try to arrest her.

Dave Bittner: Yeah.

Ben Yelin: That is a misinterpretation of two party consent laws.

Dave Bittner: By the way, hats off to her for standing her ground

Ben Yelin: Seriously.

Dave Bittner: Yeah.

Ben Yelin: I mean talk about a model journalism student.

Dave Bittner: Yes. Yes.

Ben Yelin: She certainly takes her job seriously. She was 100% in the right, at least as to the law here. And, you know, not to -- not to be biased against this esteemed congressman, but he clearly was incorrect about how two party consent law works because it only applies to secret recordings when a person has a reasonable expectation of privacy. So if you are in a private conversation, yes. It applies there. That's why, you know, Linda Tripp who was making calls from Maryland back in the day -- you probably remember this more than I do.

Dave Bittner: Yeah.

Ben Yelin: The Monica Lewinsky era. That's why she got in trouble because she was -- or potentially in trouble. She was recording these conversations over the phone where you think, you know, if you're having a one on one conversation you have a reasonable expectation of privacy.

Dave Bittner: And that's really the origin of these laws. Right? Is the telephone.

Ben Yelin: Exactly. Yeah. It all comes from the telephone as an anti wire tapping statute. So wire tapping itself as an enterprise comes from the era of telephones. You would literally tap the wires. And that's how you would record. But yeah. You don't have a reasonable expectation of privacy in public places. Certainly not while you are giving a public press conference with TV cameras around you. So that also covers the security camera as long as it is a public place. If it's recording you in your house that's something completely different. I think that would implicate a number of laws including wire tap statutes. You know, for the most part, security cameras are capturing public places where you don't have a reasonable expectation of privacy.

Dave Bittner: What about like, you know, a board room or a meeting room at a company? Let's say, you know, you're doing that in Maryland. Your watch is running. Potentially you'd have to notify everyone and get their approval?

Ben Yelin: It's probably a good idea. I mean I don't know how well tested the reasonable expectation of privacy is in a board room. Is it dependent on how many people are in the room? You know, how many people have access to it? Is it very obvious that you are being recorded? But yeah. I mean I think as a best practice it would probably be a good idea if you want to avoid violating these wire tap laws to just give notice to everybody in the room that, "Hey, in case you're donning one of those new wearables, these conversations are going to be recorded for the purpose of producing transcripts, summary transcripts."

Dave Bittner: Yeah.

Ben Yelin: So, you know, if you are admitting to any type of corporate fraud in this board room, be careful because that could end up on somebody else's device.

Dave Bittner: Yeah. I remember a few years ago there was a case. We might have even talked about it here where there was a -- there was a gentleman who was oh just being kind of a knucklehead on his motorcycle near here at like out on Route 95 or something. You know, driving down the highway popping wheelies and stuff. And he had a GoPro mounted to his helmet so he could --

Ben Yelin: Record himself breaking the law.

Dave Bittner: Right. It's always good to document your crimes.

Ben Yelin: Yes.

Dave Bittner: But --

Ben Yelin: He probably thought he looked so cool whatever he was doing.

Dave Bittner: Well, he probably did. So he got pulled over by an off duty police officer, I believe. And forgive me if I don't get every detail of this correctly, but I got the gist of it right. An off duty police officer pulled him over and the police officer drew his weapon without identifying himself as a police officer. So, you know, brandishing a weapon. And so that caused some trouble for the police officer.

Ben Yelin: Yes.

Dave Bittner: But they tried to go after the motorcycle driver because his GoPro was recording audio. The video was no problem, but the audio the police claimed ran afoul of the two party consent rule in our state. And eventually that got thrown out. But it was interesting that they tried to go after him using that.

Ben Yelin: Yes. Yeah. My how the turn tables. Right? Again reasonable expectation of privacy is the key factor here. So if you're in public as a police officer you probably don't have a reasonable expectation of privacy. If you are the one, you know, effectuating an arrest and in most states it's legal to at least in most cases videotape law enforcement so that would not run afoul of two party consent laws which is probably why that case was thrown out.

Dave Bittner: Yeah. I was at a conference oh probably about a year ago now and there was a little cocktail party after the conference and I was -- me and a couple other folks were chatting with a gentleman who had one of those little pin on recorders on his lapel.

Ben Yelin: Oh yeah. Yeah.

Dave Bittner: You know, that just records everything. And then transcribes it, summarizes it. And I just I remember one of us in the conversation was like, "Oh. So what's that thing?" He's like, "Oh. This is great. You know, it records everything. It summarizes it so every day I get a summary of all the conversations that I had." And everybody was like, "Great. Nice knowing you. I'm going to go get a drink."

Ben Yelin: I am going to go talk to somebody else.

Dave Bittner: Right. Right.

Ben Yelin: I mean I have weird thoughts about like does that record in the bathroom, you know. Like it's just we're going a little overboard with these recording devices. I can understand why it would be very useful. Like if I have one of these wearables and I'm in a meeting, you know, I don't have to turn on any AI enabled note taker. All of this could be summarized for me. That would be very helpful. If I go back to my Apple watch in two hours and say, "Hey, remind me what happened in that meeting," so great, you know. That's a great product. But it is going to run potentially in to this legal issue, especially when it comes to private interactions.

Dave Bittner: Yeah. I guess it troubles me that we're normalizing or that the big tech companies are attempting to normalize this sort of thing.

Ben Yelin: Right. And, you know, then you get in to that weird circular thing where maybe we lose our reasonable expectation of privacy eventually because we know that everybody has one of these wearables. That's why I think the reasonable expectation of privacy test, you know, is prone to failure in many circumstances because you can destroy reasonable expectation of privacy by destroying people's reasonable expectation of privacy. The government itself can play an active role and just society as these products become more available and more people have them. You start to lose that expectation that you really are in a private conversation which is not good.

Dave Bittner: Yeah. It's hard to imagine, you know, losing your reasonable expectation of privacy while in a restroom or something.

Ben Yelin: It really does. And if you don't have it there where do you have it?

Dave Bittner: Yeah. Yeah.

Ben Yelin: You know, I think the last vestige is inside a person's private home, and that's well protected, but anywhere else, especially when you're with other people these days, like this is going to be something you have to think about. Now it would still be a novelty to me if I saw somebody with Meta glasses right now, but certainly not a novelty to see somebody with an Apple watch.

Dave Bittner: No. No. Not at all. It just blends in.

Ben Yelin: Quite ubiquitous.

Dave Bittner: Yeah. All right. Well, we will have a link to that story in our show notes. My story this week is about the ninth circuit court which recently held that Copilot creates new code, not stripped copies. So this is a legal victory for AI developers. Basically this is dismissing a Digital Millennium Copyright Act claim that was brought against GitHub, Microsoft, and Open AI involving a GitHub Copilot and Open AI Codex. So the issue here was whether AI generated code violates copyright attribution requirements. Anonymous programmers alleged that Copilot reproduced their open source code without the required author credits, copyright notices, and licensing information which is known as copyright management information or CMI. So the court said that generating new code -- or rather the court distinguished generating new code from copying existing code. And they said under the DMCA removing copyright information requires an existing work from which that information was actually removed or altered. Simply generating new code without attribution doesn't necessarily qualify. What do you make of this, Ben?

Ben Yelin: I mean I feel bad for the coders. I think this is a legally correct decision under the Digital Millennium Copyright Act.

Dave Bittner: Okay.

Ben Yelin: Just because, as you said, that act does require this existing work that the information is taken from. And, you know, when you are vibe coding which I have now taken to partaking in myself you are not -- you are not directly copying somebody else's code. It is a predictive technology just like all AI is predictive. So it has been trained on millions of lines of code and will predict what the next line of code should be based on what it's been trained on.

Dave Bittner: Yeah.

Ben Yelin: And that is not directly copying somebody else's code. Now if I'm a coder like that makes me really angry. The toil that I've put in to my labor, you know, over several years, decades, somebody can now do just by putting in a prompt to Chat GPT being like, "I want to create a new app that has a really cool interface." I can see why that would be very frustrating, but it's not a violation of the DMCA. Now one thing that this article notes is that this ruling doesn't establish that AI generated code can't infringe copyright law at least under common law. So the output of kind of vibe coded whatever, vibe coded content, may face different legal consequences under the traditional copyright statutes and common law that we have.

Dave Bittner: Yeah.

Ben Yelin: If there is a scenario in which an AI system retrieves and reproduces explicit stored content, you know, that's going to be a different story entirely. But for this scenario where yes it's based on existing code, but it's not expropriating that code directly I think this is the legally correct decision even though as a coder, like I said, I would be very annoyed by this.

Dave Bittner: Yeah. Have you seen lately there's been several efforts, some highly publicized, of I think in one case somebody basically recreated the Adobe Creative Suite in a new language? In other words, Adobe, I believe, codes using C ++ to create all their apps. This person put his large language model, his coding agent, at the task of reproducing these using Rust which is a different language. So the point being that there would be no shared code. There'd be no borrowed code. They were recreating the apps using all new code and this person has created functioning, you know, version 1.0 of all of the apps in the Creative Suite. And so far Adobe hasn't responded to this.

Ben Yelin: Yeah. See that I'm not sure of because let's say I reproduced a great work of fiction, but did it in a different language. Wouldn't that be by definition a copyright violation?

Dave Bittner: Well, that's a good -- is it? It's a good question.

Ben Yelin: I mean I think it is. Right?

Dave Bittner: But none of the words are the same.

Ben Yelin: None of the words are the same, but it's still the same copyrighted -- like it's still the exact creative output of the author as it would be in this scenario.

Dave Bittner: Yeah. I think there was a -- there's a prior case that has to do like that's computer related that has to do with back in the day when folks were making IBM PC clones one of the clone makers created a functionally compatible bios which is kind of the underlying, you know, software that helps the computer run. They created a functionally identical or compatible bios from whole cloth. In other words, they set an engineer on doing this, didn't let them look at the thing they were trying to copy, just described what it needed to do, and it was found to be legal because it was created from whole cloth. So evidently what I've seen or people pointing to that case in these cases with the Adobe thing where there's some prior case law saying if you make it up in a new language or don't refer to the original you can get away with it. We'll see.

Ben Yelin: Yeah. I mean that just seems wrong to me just because we wouldn't accept that as any other -- in copyright law in any other context because a new language still it's still appropriating the original creative work. And it still deprives the creator of the fruits of that labor. You know, and I think there's a big distinction between what vibe coding is which is basing new code off training of billions of lines of existing code than just translating somebody's created code in to a new coding language. I mean to me that's a difference of category. So maybe that's not how the law will end up seeing it, but I don't know. I mean don't you think it -- don't you think there's a distinction there?

Dave Bittner: Oh. I do. Absolutely. And I just wonder what sort of legal corner have we been backed in to here. Right? As I said, with the prior ruling on the bios case what does that -- how does that implicate what's going on here? Adobe has been very quiet, but you can bet that their legal team are putting in the extra hours to try to figure out what do we do to protect ourselves. I mean I've seen some folks say this could be the end of commercial software. Right? If this is legal -- because the same people who did this to Adobe did it to Microsoft. They recreated the whole Microsoft suite.

Ben Yelin: Just in a different language?

Dave Bittner: Yeah. The same way. Functionally identical. And there have -- I think it's worth pointing out that there have been open source competitors to the Adobe suite for decades for people who don't want to have to pay Adobe for, you know, the -- to use their tools. But they've always been distinct.

Ben Yelin: Right. That's --

Dave Bittner: And this is not.

Ben Yelin: That's why it can be open source is that it's distinct in some way.

Dave Bittner: Can they come after them for some kind of like a trade dress or look and feel kind of thing? Is that a play here?

Ben Yelin: I mean I think the way the law works is it has to be like any sort of derivation generally ends up being a fair use. Any sort of derivation or inspiration. But if it's a direct expropriation of the original content then that is the copyright violation.

Dave Bittner: Right.

Ben Yelin: So when you have one of those open source -- when you have open source software where it's similar to Adobe, but not identical then I don't think that it becomes a copyright violation just because it's not, you know -- it's taking inspiration from somebody else's coding work, but it's not copying somebody else's coding work.

Dave Bittner: Yeah.

Ben Yelin: You know, maybe it's a distinction without a difference, but to me that even seems different than your original Adobe scenario where it is using the exact same coding. It's just in a different coding language.

Dave Bittner: Right.

Ben Yelin: Like I think you should have to make some changes to it. You have to input some of your own creative work in order to avoid a copyright violation. That's the way I think the law should be.

Dave Bittner: Well, but I mean yes and I think one of the interesting outcomes of this is that the AI created version cannot be copyrighted. Right? For that very reason.

Ben Yelin: Right.

Dave Bittner: Because it's not -- no humans did anything.

Ben Yelin: I mean that's a whole other can of worms.

Dave Bittner: Yeah.

Ben Yelin: If we change liability law to hold AI accountable one way or another, especially, you know, in this new age of agentic AI, maybe that type of thing changes if you actually have a copyright violation. But it's an entirely separate question. Like how do you hold AI accountable for committing some type of legal wrong?

Dave Bittner: Yeah.

Ben Yelin: You know. There's the question of that whole can of worms and then there's the legal wrong itself which is I think what we're referring to here. But yeah. I mean it's going to get super confusing when, you know, maybe you get a closer case than the one at issue here where you don't have exact duplication, but you know it's a near verbatim copying. And AI is responsible for it. You know, the frontier models who develop AI tool will say like, "Well, we had no idea that it would do this." And, you know, right now they'd probably be shielded from liability in that context.

Dave Bittner: Well, I wonder also what this means. Like if I could do this in the privacy of my own home, right, you know, I'm sitting home. I need to do something that's Photoshop like. I don't own Photoshop. And I go to my LLM, you know. Let's say 5/10 years from now when they're an order of magnitude more powerful than they are today. And I say to it, "You know, I really need some -- I need a Photoshop compatible app."

Ben Yelin: Right.

Dave Bittner: Whip that up for me. And it says, "No problem. Give me five minutes."

Ben Yelin: Five minutes? Could be five seconds.

Dave Bittner: Right. Right.

Ben Yelin: Yeah.

Dave Bittner: And it does so. And it's no big deal. Right? What world does that put us in where anything that's created by software can be just spun up spontaneously? How do you put guardrails on that?

Ben Yelin: Yeah. I mean the ultimate issue here is that it's going to disincentivize people from going through the toil of creative work, of creating new code.

Dave Bittner: Right.

Ben Yelin: Now people have probably said this about every type of technology that, you know, once some new technology becomes available it makes somebody else's old work obsolete. But I do think that's a major issue in the AI age, that maybe people won't get -- you know, I think a lot of coding positions at organizations are going to be cut just because this is a skill that AI and, you know, especially Claude in particular has really perfected.

Dave Bittner: Yeah.

Ben Yelin: And it's just unfortunate if you've made your career, you know, learning these very complicated coding languages, which I never learned which is why I feel bad doing my vibe coding, but --

Dave Bittner: It sure is fun though.

Ben Yelin: It sure is fun. And this is the day. This is the age that we live in.

Dave Bittner: No. I think you're absolutely right. So I guess the bottom line here is that AI developers have gained protection against one category of copyright claims, but there's still legal obligations around training data and code reproduction and attribution and open source licenses that remain unsettled.

Ben Yelin: Yes. This is just the beginning, I think, of a new body of law on AI and intellectual property. You know, I don't think we're anywhere near settling some of these broader issues. And this is just related to coding. I mean there's going to be a whole body of law that develops for other creative work. You know, art, music, creative writing. Like it's just it is something that's very unsettled. And, you know, people who are in the creative industry I think are rightfully terrified that they're no longer going to enjoy the fruits of their labor if AI can produce something that approximates their own creative work. It can do so very easily. And it can do so in a way that's not easily traceable. That really puts people in creative industries at a disadvantage. So it's not just coders who are going to have to wrestle with this problem. I think it's anybody who does any sort of creative work.

Dave Bittner: I have long been of the opinion that we need a fresh look at copyright law here in the U.S. I think there's some -- some things have evolved in ways that were not copyright's original intentions and, you know, big companies have used their influence to make copyright protection lean their way. Famously for decades any time Mickey Mouse was about to go in to the public domain copyright got extended.

Ben Yelin: Right. He finally is in the public domain though.

Dave Bittner: Right. Right. Yeah. Disney finally gave up that fight. They saw which way the wind was blowing. But, you know, I think maybe if this is the thing that causes a fresh look at AI maybe that's not such a bad thing. I don't know. My personal opinion, but we'll see how it plays out.

Ben Yelin: Yeah. I know. It's a very interesting story. I think this is just going to be the very beginning of a new body of law that develops on this.

Dave Bittner: Yeah. All right. We'll have a link to that story in the show notes. Let's take a quick break here to hear from our show sponsors. We'll be right back after these messages. [ Music ] And I recently had the pleasure of speaking with Danny Manimbo. He is ISO and AI practice leader at Schellman. And we're discussing why companies waiting for AI regulatory clarity may already be behind. Here's my conversation with Danny Manimbo. [ Music ]

Danny Manimbo: 2026 was supposed to be the year of AI regulation. Right? We were -- I'm in Denver, Colorado. We were supposed to have the Colorado AI Act had gone live at the end of June. You saw that potentially. That was -- that was a, you know, basically taken away. Right? They did away with the Colorado AI Act completely. It got replaced by what they referred to as an automated decision making framework and ADMT framework that goes in to effect in January of next year. Outside of that particular update with the Colorado AI Act there's been kind of a long saga. Our governor originally signed that in May of 2024. He signed it with what he said was a lot of hesitation. You know, wasn't overly thrilled with the contents of the bill or how they would make rule making, etcetera, for organizations to comply. Got a lot of pressures from business. So it was originally supposed to go in to effect in February of this year. They pushed it back. And just, you know, it never even got liked before it was totally abolished. So that was kind of the major comprehensive AI regulation in the U.S, at least the first of -- first of its kind and, you know, kind of the first one to go in to effect there that would impact a lot of organizations. And of course internationally all the talk's been about the EU AI Act and they had it was really all in the month of May that this happened with Colorado and what we saw with the EU AI Act as a result of the AI digital omnibus. They moved back their high risk AI system obligations which were supposed to go in to effect in August of 2026. And pushed those back, you know, in to 2027 and potentially 2028 depending on the, you know -- whether it's a standalone or an integrated AI system. So it's just been a lot of shifting timelines. So I think, you know, organizations that are trying to peg their compliance programs to specific regulations which we don't advise for this very reason are kind of like scratching their heads.

Dave Bittner: Is it fair to say that there are lots of organizations out there who are waiting for regulators to provide a clear road map before they take action here?

Danny Manimbo: They may be. Right? Because we certainly saw -- you know, we've seen an uptick in our compliance services, you know our AI insurance services, if you will, which is, you know, primarily ISO 42,001. We've done some news day IRMF assessments and this year expanding more in to the agentic offering with AI UC 1. But certainly this year we have had specific inquiries around we need to be certified by, you know, August which is the EU high risk obligation deadline. So some of that conversation obviously has changed with a lot of the headlines, but you know a lot of what we're advising clients is look, you know, regulation or not, AI risk remains. And I think the market is becoming increasingly more and more aware of what those risks are, the types of questions to ask, because those have changed a lot since 2023 when a lot of these governance frameworks were published in the news day IRMF and ISO 42,001. It kind of showed what trustworthy AI could look like, but, you know, nobody was really talking about agents and the agentic AI in 2023/2024. So, you know, those questions that are being asked have really changed and, you know, as a result placed a lot more pressure on companies that are providing AI solutions or developing AI solutions to have, you know, answers in place and be able to have trustworthy AI programs.

Dave Bittner: What does good governance really look like? You know, a good governance framework, what does it actually consist of in the real world?

Danny Manimbo: You know, some of the basics. I think a lot of folks can overthink this, but, you know, a lot of the basics you're even looking for on good security programs as well. You know, you have defined ownership. Right? Policies. System inventories. Right? You can't -- you can't govern what you can't see. And this concept of shadow IT or shadow AI basically as unauthorized or unnoticed use cases of AI across organizations is certainly present these days with democratization of what we're seeing with all the GPTs and things like that that you can have. You know, performing risk assessments, having management oversight. I mean that last one I can't underscore because if AI governance or any governance program really is delegated down to a compliance manager, you know maybe down from CISO or an AI officer down to a compliance manager and it just sits with them squarely, I mean that's really where these programs they go to die. Right? Because they really do need ownership not only upwards, but laterally as well to other, you know, business units within an organization that could be -- that could be data science. That could be legal. That could be a product team, developers, you know. Everyone's playing a role.

Dave Bittner: When you work with organizations that are getting this right what are they doing differently from companies that are say out there, you know, checking that compliance box?

Danny Manimbo: Yeah. It's a great question because there's certainly -- there's certainly a lot of that out there. But, you know, I think the companies that are getting this right we can tell pretty quickly. You know, just about everybody has an AI policy. Right? But whether or not that policy's effected. You know, are they actually performing what you hear the term impact assessments? You know, this really came out with GDPR in 2018 with privacy, but you know with AI it's just as well. It's in the EU AI Act. It was in our original Colorado AI Act as well. And that's really more than just a security risk assessment. That's really kind of doing almost a traditional risk assessment might be an inside looking out approach. This is really an outside looking in approach. You know, what are the potential benefits and harms of our AI system to, you know, individuals, to society? You know, what does misuse or abuse of our system look like? What are some safety considerations? You know, if you think about a lot of the use cases of AI, anything from, you know, autonomous driving, you know, these are things that are going to have real human safety, you know, requirements that we need to -- or obligations that you need to consider. The algorithms, any upstream reliances that you may have on the model algorithm providers, these are all things. What does quality data look like for you? Right? Because we know that it doesn't matter how powerful your algorithm is or how -- you know, how the latest and greatest model that you have. If these things are not operating and trained on the right data they're not going to make the right decisions. So I think it's thinking through a lot of that. And probably the last way I'll answer that question is I like to call it, you know, the fingerprints test. You know, show me an example of, you know, a decision being potentially overturned because the AI governance committee stepped in and said, "Look. We have to turn off that feature because, you know, it may, you know, run in conflict with something that the EU AI Act calls out or we may need to step in and do a human override on this -- on this decision because if you're just letting the AI kind of run and there's no human in the loop, human on the loop, and everything, you know -- everyone's just kind of quote unquote trusting the AI, that's where things can start to go off the rails.

Dave Bittner: Looking at the procurement side of things, are organizations beginning to see AI governance questions showing up during vendor assessments or procurement reviews?

Danny Manimbo: Absolutely. And this it took a little bit for this to occur. Right? Because we kind of think back to the last couple of years. I mentioned 2023. That was when NIST AI RMF and ISO 42,001 were published. 2024 the market is just starting to get introduced to these standards. And it's all fairly new. Certification bodies like Schellman were still getting accredited. So there wasn't really anybody certified in 42,001 until the second half or even the fourth quarter of 2024. So as far as kind of setting the tone for the market in terms of what trustworthy AI look like it's all fairly new. And people were kind of looking around. Who was going to do it first? And it was a lot of the enterprises kind of setting the tone. Entering 2025 I think more procurement teams, supplier due diligence, you know, security questionnaires started to incorporate AI specific lines of questioning because they didn't really know the types of questions to ask prior to that, but they had now established standards in, you know, things like 42,001 where oh okay well this is what ISO says a trustworthy AI management system looks like. Are you doing these things? Much like they've done with 27,001 for information security throughout the year. So we have started to see procurement start to ask a lot of these questions and, as I mentioned before, you know with kind of this evolution of the technology towards agentic and autonomous non deterministic systems it's really changed the narrative there because that really wasn't top of mind when 42,001 was published three years ago. So it's kind of changing there quite a bit as far as the types of questions that are being asked during those procurement reviews.

Dave Bittner: What's your practical advice for folks out there who, you know, they're either just starting this journey or maybe they're a little bit down the pathway, but they still feel like they're a little bit behind? Any words of wisdom here?

Danny Manimbo: Yeah. Absolutely. I mean don't start from scratch. Likely, you know, organizations already have some type of compliance program in place. Right? They may have obligations for information security or privacy. And kind of one of the mistakes that we see organizations make is well we're going to establish our AI governance over here. Going to move my hands which you can't see. But we're going to have security and privacy, you know, over here. And they're going to be very siloed and the compliance programs aren't going to talk to one another whereas if you look at something like ISO 42,001 which, by the way, is a great place to start because it's a intentionally a very broad framework that is intended to apply to people who are actually developing the AI models to folks who are just using, you know -- incorporating AI in to their SAS platforms to folks who are just using AI internally within their organization. So it's intended to really apply to any organization and provide them with the scaffolding of establishing a trustworthy management system. It's very aligned with a lot of the main themes and regulations that we're seeing. So starting with something, you know, that's concrete, that's proven, that's out there, that's internationally accepted I mean I think would help you get confidence that you're moving in the right direction as far as implementing any AI governance program. And of course you want to get the right buy in across the organization, to my point earlier. You can't do this in a silo. You've got to get buy in from leadership and the right eyes and overall accountability structures established.

Dave Bittner: Is AI fundamentally different here? I mean if an organization already has a mature cybersecurity and compliance program does AI governance fit within those existing structures or is this a different beast?

Danny Manimbo: I think it fits within because if you look at ISO 42,001 the standard specifically notes that it is intended to integrate with a security program for ISO 27,001, a privacy program for ISO 27701 or even a quality program for 9,001. The latter is probably not talked about enough because the EU AI Act actually requires a QMS, a quality management system, because going back to my point earlier about the importance of data quality and being very process driven I mean customer confidence driven standard, you know, it's extremely important. But I would say there are a lot of unique risks. While there are technology risks, etcetera, there are a lot of unique risks to AI which are not the established intent of these cybersecurity frameworks that we've seen out there for years. Like I said, I've mentioned about 27,001. It's not intended to cover or map or replace anything in 27,001. It's really meant to be its own thing in 42,001. Same thing with SOC 2. I mean that was -- that was drafted a long time ago and it hasn't -- it didn't incorporate, you know, certain AI risks like, you know, bias and fairness and safety and a lot of those maybe non technical risks that you may see covered in a cybersecurity world.

Dave Bittner: Ben, what do you think? People doing a lot of catching up here?

Ben Yelin: Yeah. I mean the last thing you're going to get in this country is AI regulatory clarity.

Dave Bittner: Right.

Ben Yelin: You might get some of that in the European Union with the EU AI Act, although there are certain provisions of that that are being contested in many ways and it's not an entirely crystal clear framework. You know, it is a risk based framework and it's not always clear what level of risk applies to what use case. But that's way better than the wild west that we have here where there's a distinct lack of federal regulation and you are going to have to live in the world if you operate in the United States with the absence of any regulatory clarity.

Dave Bittner: Yeah. State by state. Right?

Ben Yelin: Absolutely. Yeah.

Dave Bittner: My heart goes out to folks like Danny here who have to help - have to try to make sense of this, bring order to that chaos, and guide their customers, their clients who they're advising, on what to do. It's a --

Ben Yelin: It's making a lot of compliance officers and lawyers very red. So at least there's a plus side to it.

Dave Bittner: Well, I hope Danny's doing well. And again we thank Danny Manimbo from Schellman for joining us. We do appreciate him taking the time. [ Music ] And that is "Caveat" brought to you by N2K CyberWire. We'd love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like our show, please share a rating and review in your favorite podcast app. Please also fill out the survey in the show notes or send an email to caveat@n2k.com. This episode is produced by Liz Stokes. Our executive producer is Jennifer Eiben. The show is mixed by Tre Hester. Peter Kilpe is our publisher. I'm Dave Bittner.

Ben Yelin: And I'm Ben Yelin.

Dave Bittner: Thanks for listening.