AI Watermarking: Compliance Theater or Real Provenance Tool

0:08Aaron: Good morning, good evening, where you are, and welcome back to the Enterprise AI Show. This is your host, Aaron. And if you're listening to this as an audio podcast, this should feel like just another show. If you're catching us on video, mainly through our YouTube channel these days, this might be the biggest change in our 1000 plus episodes. As of today, we are converting the entire workflow from recording to editing all to video first. And we still have some things to work out like me forgetting this transition and then coming to the massive studio down here at the beach without my lights, the good camera or the teleprompter for this intro, but the show must go on and things will continue to improve over the next few shows.

0:47Aaron: Now, moving on to today's show, Brian, Brandon, and I, we talk about AI watermarking and whether it matters long-term. There are regulatory concerns, privacy concerns, and of course, another front in the never-ending cat and mouse game for those that just don't want this. And all of that is coming up right after this break.

1:11Today's show is sponsored by Nasuni. There's a growing gap in AI right now between what's possible in theory and what successfully works at scale inside an enterprise. The difference comes down to unstructured file data. Many AI initiatives struggle because the file data they depend on is scattered, unstructured, and disconnected from where and how work actually happens. Nasuni changes that. It brings your unstructured file data into a single, secure foundation so AI, both generative and agentic, can access it with the context, governance, and performance it needs in production. Bring AI to where your unstructured data lives.

1:45See what it takes to activate your data for AI and request a demo at nasuni.com slash AI.

1:54Aaron: And we're back. And we have the gang all here again this week. And Brian is either getting ready to go golfing, returning from golfing, or hasn't left his chair in a couple of days. We're not going to tell you which one that is. If you're on the video, you'll know what I mean. We have a topic for today. And actually, this is one I'm kind of sort of surprised that it hasn't happened sooner or hasn't kind of made the mainstream press sooner. And that's kind of this AI watermarking that's going on. It already was going on a little bit behind the scenes with Google and then Anthropic notoriously has come out with that.

2:29Aaron: And then, oh, by the way, the watermarking got cracked within 24 hours and there's already clod strippers out there. I don't know if that's a good term, but anyway. So guys, let's talk about this for a little bit. The larger aspect here is like for the EU, there's a certain EU AI Act. It's the EU AI Act Transparency Code is the one in particular. But there's that, there's GDPR, there's all these other things going on. And everyone kind of wants to go, hey, was this AI? Was this not AI? So this is something we all saw coming in the industry. But is it good? Is it bad? Where's it going to go? Let's talk about it today.

3:09Aaron: So Brendan, you want to kick this one off today? Yeah, my take on the watermarketing thing is I think it's fine if people want to watermark it. But to me, it doesn't matter to me either way. I'm okay if things are generated with AI that people send me, whether it's long written text or if it's an email or whatever. All I really care about is, is it good? Whatever I'm reading, does it help me do what I want to do? So I think there's just a lot of, I don't know, pearl clutching amongst people that write. Like I always think the writers and the press and like authors are always just every word I write is chosen with this incredible thought and preference.

3:47And how dare that someone think that my writing, even if I just used it for editing, was done by AI. So it's one of these things where it's like, I think it's fine if it's watermarked. I don't think it necessarily like helps anything in any way. Like I'm not going to worry about it. And I just assume that most everything that I'm reading these days has passed through some type of AI filter in some way. And I almost want to put in my own email, own disclaimers, like, yeah, if you're getting something from me, probably went through AI. Like, I probably at least did an edit. And certainly if it's at all important, right?

4:18Maybe my fantasy football Slack messages, like, maybe they don't go through it, but everything else does. So, you know, it's one of these things where I don't know if it really solves a big problem, but if it makes people feel better that it's watermarked, it's so be it. What do you think, Brian? For somebody who hasn't read a lot of the articles on this, what are the basics of this? Why are they doing this? Who are they trying to protect? What are they trying to protect against? Yeah, the short version is basically it's just regulation. The EU has basically said that, listen, if you produce something with AI, it needs to be clear that it was produced with AI.

4:49So that's where the watermark comes from. And then for those that haven't been following it, we'll make a really long topic that's kind of boring short. We'll just say the way it does it is it's not actually like putting characters in there. It uses some math that says when you select certain words, when it's picking the token, the word that actually, the AI will basically remember, be like, I won't use a purely random number here to generate this next word. I'll kind of know what it is. And then if you run a tool, it can look and say, oh, I see these words were all picked with these numbers. I can kind of tell that this was generated from AI.

5:23So the need is just today is 100 % EU mandated. So it's compliance, I guess, is ultimately. And then Anthropic has made the case that like, or made the decision that they're just going to do it across the board. They're not going to try to do it just for the EU and not for others. And as far as, I don't know, do we need it? That's the open question, right? I think people would like to know maybe if things were used from AI, but is it really important? I don't know. That's maybe a personal, maybe that's a question for you, Brian. Do you care? Do you care if somebody sends you AI generated email?

5:52No, not really. It's most of the time, the people who do it and send it, like you can tell pretty quickly because it's got a bunch of em dashes in it. Like that's always to me the telltale thing. I guess I'm trying to ultimately figure out like, is this another sort of EU effort to like take a bigger problem. Like if they're concerned about, Hey, we want to, we want to avoid all these like false things getting across society and society getting manipulated and stuff. Okay. That's fine. But they make a big deal and they're like, okay, we're going to just scratch the surface. And the remedy for this is going to be a small fine that the U S companies intentionally ignore for a while.

6:26And then at some point they deal with it. And ultimately it just creates a worse scenario for like the European citizens. Like I, this is where I'm like, we have a long history of, we have a long history of pop-up blocker. Yeah. The pop-ups GDPR and other stuff where you're like, God, that is a lot of work and it costs a lot of money and it benefited nobody in any way. Like I I'm trying to get why this is happening. I guess it does feel like if this was mandated by an EU thing to prevent whatever, like it's either like the horse is out of the barn or you scratch the surface, but like, what did this really do?

6:59Aaron: Here's my biggest thing too, is I don't know that any of us are the target audience. Because I think the target audience is people who actually care like, hey, was this, did somebody like really generate this or is this AI generated? And I don't think all three of us, none of us care. And if anything, we probably work in a lot of AI generated stuff. We work with a lot of AI generated stuff. We receive a lot of AI generated stuff. We're a little numb to it. And honestly, if it's better than what you would send me, what you created, I'd rather be AI generated anyway. so i don't care i like i just flat out don't care about yeah this feels like a this feels like a very small percentage of your life like we all get crap loads of ads those are already ai generated you get crap loads of emails those are all ai generated or at least they were swapped to begin with you get tons of robo calls and so on and so forth it's yeah if you interact with some person you're gonna you're gonna give it a little bit of thought to okay but at the end of the day like brandon said you're gonna be like okay was this useful helpful intelligent unintelligent whatever I guess my only curiosity is like, what tax has this added to doing this?

8:07Because if they have to do additional computation, this obviously costs them something that gets passed along to people. Is this a 1 % tax? Is this a 7 % tax? Or like, what does this really do?

8:18Aaron: No. This means the only interesting piece. Yeah, this feels to me like the companies will meet the letter of the law and nothing more, and everything will be super easily circumvented, and everyone will just move on with their lives. I really think this will be a non-issue except for those that are diehard for it. I feel like if Anthropic immediately just said, hey, we're going to do this for everybody, not just for the EU, it must not be that big a deal. because you look at every other EU mandated thing and the companies, they go to the mattresses over fighting this thing. And so it must not cost them nearly anything.

8:57And maybe this is one of these anthropics, like we're just going to say it's everywhere so that we just make a standard operating procedure. We move on and maybe this will catch up open AI somehow and we'll get a little bit of advantage over them in Europe or something like that. I think it's going to be frustrating for all the students going back to college. They're going to be like, I was telling them, mom was telling my son, I was like, listen, just so you know, everything's watermarked now. So you better watch out when you turn it in the professor, should they be smart enough? They can, they can probably figure it out, which would be, which would be funny.

9:25Yeah.

9:26Aaron: Okay. Super interesting point there. My, with my youngest, we have a lot of conversations about it because at her university, they're very anti AI and all the professors literally are like, all the policies are, if you get caught generating anything with AI, you can fail the class kind of thing. Like pretty extreme about it. And it's interesting too. Like I have Grammarly, I use Grammarly all the time. I love Grammarly. Grammarly now has an AI detector in Google Docs. And it's so funny to see literally everything you get from anybody else, like any other vendor or things you generate or things that people generate to you.

10:01Aaron: The AI detector goes off on every single document I use every single day. And so I just had to turn it off. Like it became that useless because it was like, yes, we're all of course generating everything. But like my daughter, she was the exact opposite. She's like, I cannot do it. I can't touch it. Or of course, it's like, hey, you might run it and then you might, there's the answer and then you retype it in kind of thing. But yeah, it's interesting to me. I can see this being a big deal in education. I can see certain industries where this could be a big deal. Well, the other thing professors are doing, and I know we're getting off topic, is the professors are, they're going, hey, they're basically assuming that you're going to cheat for everything if you're allowed to.

10:45And so they've gotten smart and they said, okay, this is what we're going to do. 95 % of the class is worth 50%. That stuff can be remote. It can be in person. It can be whatever. It can be test. You're probably going to cheat on that. So I'm going to give you the first 50 % for free. And then the last 50 % is an exam is no outside tools has to be in person, cheat proof. and took the simplest way for them, I think, to figure out, okay, you might've cheated for the first 50%, but you can't cheat for the second 50%. So we'll see if you paid attention. Yeah. But it seems reasonable. I don't know, to me, I think it's like, this is the part where the education always gets a little confused.

11:18It's like, it's different if you're trying to see if someone has mastered some subject, right? Like adding or subtracting, something simple, right? Like, obviously we all like learn how to do that, but then most of us just use a spreadsheet or whatever going forward, right? And so I always think that's just, when I think about education, that's what I always think about. Yeah, there's a little different. You do have to have some in-person exams and make sure someone demonstrates mastery. But at the same time, I just go back to, I know it's been quote unquote a long time since I went to college, but it's like they would always say have someone read, have a friend read your essay, right?

11:48Like approve, like that was like a very common thing was like, have them read it. You'll probably miss some typos or whatever and or whatever. Peer review editing, it's been around since the dawn of writing. So the fact that you can't allow the best peer review editor you have access to review your document and help you like, oh, you misspelled something, used the wrong word here. Here's a way to say something a little bit more clearly. To me, it's just like a miss. It's just like a complete miss. Where it's going to get really interesting in the enterprise is around procurement. So people are going to send out RFPs and RFQs and stuff like that.

12:21And they're going to put some sort of language in there that says you can't respond to this with AI. And then, of course, everybody's still going to respond to it with AI because they always give you like three days notice right at the end of a quarter or something like that. And it's going to be interesting, especially like in the government where they have these lawsuits all the time about you guys didn't give me fair shake and what they did. If somebody being like, oh, I did it without AI was my way of getting around having lost that bid. I think there's going to end up being a lot of lawsuits and a lot of paper trails about RFPs and RFQs and how you responded to certain stuff that involves actual money that I don't think anybody's thinking about what those downstream costs are going to look like.

12:59Because those things will come up. Somebody's going to flag something as like, you did it, and they're not going to let you get in. And then they're going to be like, what happened if everybody got flagged? Now what do we do? Your contract has been watermarked, therefore it's been null and void. It's funny you say that because I feel like it's already leveled the playing field so much because I think everyone that's in business is like just a regular employee. You've probably been given some 20 page document and asked to sign it, whether you work somewhere or you're doing a contract and you don't have a lawyer.

13:26You can't pay a lawyer to review it. It's just not feasible. And the fact that you can throw this stuff in the eye and I ask it all the time, read this over and tell me what are things I should be concerned about. And it's amazing what it will highlight. And then it will give you a chance to be like, okay, we'll draft some changes to this and I'll send it back. Whereas before your only choice was either sign it or just not. And I think it's, I don't know, it has equalized things a lot. It's almost allowed us to, we could be more real about this. Don't send us 20 page documents if you don't want to engage with an AI.

13:59If you want to do a one page agreement written in simple English that we can all understand, fine. Let's almost back it out. Let's go back and be like, let's just be clear. If not, then that's fine. Your lawyer can talk to my AI or my AI can talk to your AI until we reach some agreement, which seems like where everything is going. And that's fine. At the end, if it's all watermarked, fine with me.

14:20Aaron: and i almost liken it to there's almost going to be this default mode of like people see you don't see it as much now but when people want email on phones and it would be like sent from my iphone and it'd be like oh excuse the typos or excuse the brevity or all these other things and like you could tell the difference between an email typed on a phone and an email typed sitting at your laptop i almost see this scenario in which it's oh i i had ai review this or i didn't have AI review this. And then some people almost have different levels of expectations of, oh, it's just him. He didn't have his AI review it.

14:54Aaron: So I'll go easier on him kind of thing. Or you have a certain expectation level based off of that. Or I just see a world in which we like, honestly, almost start to self-disclose stuff like this. I don't know. I just work in an industry as I assume everything I read is AI generated or AI assisted or AI reviewed. And by the way, everything. Could you even go further? Could you even go almost to the other side of this and be like, for all business communications, I'm talking about emails, contracts, status reports. It's like, I almost prefer that like, I want this to be AI generated, right? Like I want it to be, and I like, you almost just give everyone a format.

15:32I want it to be very concise. I want it to be like this. I want the emails. I want, because you think about all the emails we send and read and write. And sometimes I just, I don't know, even my own writing, I'm just like, oh, this is too long. And it's just like, it's great if everybody's emails, like, it's almost like if the net thing is like everybody's emails are a lot shorter and to the point, it's a huge one for society. I don't know how much that's worth, but it'd be great. I never wanted, like, I see a three page email or sorry, a three or four paragraph email. I'm just like, I'm putting that in AI.

15:59Like, I don't want to read that whole thing. What are the three things I need to know about? So maybe that's the way we should go is not only should it be watermarked, if it isn't watermarked, I'm not, I'm not even going to read it. It means you didn't make it brief enough for me.

16:10Aaron: yeah i think that went the opposite of where they wanted us to go with that but i love it i was fully i was fully expecting this thing when brandon was explaining to me that this was all about distilling and it has nothing to do with distilling so again i learned something no i will say this though everybody at the wall street journal and the new york times the new york media class i think they're very upset about this they're upset with the ai stuff and they would they're very they'd be upset that everyone's using ai to write so we don't have anybody on the show to represent that point of view, but the literary class is not happy with all this AI writing.

16:39So I guess you have to listen to a different Derek Thompson podcast to hear about why the world's going to hell because we're using AI for writing.

16:48Aaron: Yeah, but at least we submit our biases right up front. This is a biased podcast and that's the way it is. Love it. Brian, you want to close this out, man? Yeah, sure. Absolutely. Folks, hopefully this was interesting to you. Hopefully more than one of you other than me learned something in this one. And guys, thanks for the time today, folks, whether you watermark or don't watermark, we're not going to judge you. We just hope you give us a five-star review on all the places that you get your podcast, whether that's Spotify or Google or Apple or wherever it might be, or even on YouTube. Give us a thumbs up on YouTube.

17:17So with that, we'll wrap it up and we'll talk to you next week. Thanks for listening. Check us out at theenterpriseaishow.com for past shows, newsletters, and all things enterprise AI.