In-Car Surveillance is Coming | Diet TBPN
29 Apr 2026 · 17 min
In this episode, the hosts discuss the looming U.S. mandate for passive in‑car alcohol detection technology slated for around 2027, examining the policy background, technical options (breath sensors, fingerprint scans, driver‑facing cameras), and the massive scale of American driving that makes even 99.9% accuracy generate tens of millions of false positives. They explore privacy and surveillance concerns, libertarian backlash, conspiracy theories about remote kill‑switches, and the potential impact on classic‑car collectors who may see a premium for non‑surveilled vehicles. The conversation then shifts to a quirky AI issue: the GPT‑5.5 coding agent’s “goblin mode” prompt that forces the model to avoid mentioning creatures, illustrating emergent model behavior. They also cover AI‑generated video mashups that blend Star Wars and Pawn Stars footage, discussing IP and the future of tool‑driven VFX, and finish with a brief roundup of live‑blogged courtroom drama and podcast recommendations. The tone is conversational, peppered with humor, skepticism, and occasional optimism.
Chapters
I wrote about this in-car surveillance thing. It's going viral. It's sort of fake news. We gotta sort of truth zone it, but also it's coming and it might not matter because if you get in a robotaxi, there's definitely gonna be a camera on you. But the fear is centered around this idea that there will be a requirement from the federal government that every new car sold in the United States will be required by law to have technology, Uh-oh. That puts constant surveillance on the driver. And this is happening sooner than you think, by 2027. That's just 12 months away. AI in your car will determine if you're sober and fit to drive, automatically turning off the vehicle.
It's almost May, by the way.
I know. We're what, 8 months away, I guess, from 2027? This is the real AI 2027 problem. This is a big deal.
We are having audio issues. It's most likely a nation-state. Potentially possible we are under attack. We are working on it.
It's like watching a dubbed foreign film. Well, uh, maybe we can play this video of what happened the last time America tried to pull back on driving a car while intoxicated and some of the, some of the response, some of the backlash from any attempt to restrict drinking and driving here is viewed. Can we pull this up? Okay, we're working on it.
Anyway, production team is being humbled this morning.
Humbled. Uh, anyway, let me read through some of my take and then we'll play that funny video. So, sounds scary. Yeah, let's play this video. Drinking and driving here is viewed by some as downright undemocratic. It's kind of getting common when a fella can't put in a hard day's work, put in 11, 12 hours a day, and then get in your truck and at least drink one or two beers.
They're making laws where you can't drink when you want to. You have to wear a seatbelt when you're driving.
Pretty soon we're gonna be communist country. Communist country.
When was that? Was that the—
Apparently that's real. I've seen that video before. I always thought it was fake. I didn't realize that that was real. That feels like '80s, '70s, something like that. '80s. '80s, okay. Well, there was backlash then and there's backlash now. Some of it's a little bit overblown. This sounds very scary, you know, constant surveillance. The real, the real crazy version is camera that's watching you at all times. The government, the police, and the automaker can just turn off your car whenever they want. That sounds bad. What's actually being proposed? What's actually on the timetable? So the rationale for this is good. And I think most people would agree that it is a reasonable thing to do to try and curb alcohol-related vehicle accidents, drunk driving. There's more than 10,000 alcohol-related deaths each year on US roads. In the United States. That's a ton. Anything we can do to stop that, reduce that, totally worth working on, totally worth pursuing, as long as it doesn't violate a whole bunch of other liberties or create more problems than it solves, right? We want net positive impact here. Technology is getting better at detecting intoxication, and it's getting a lot cheaper. So why not just ship drunk driver detection system with every car from the factory? That's the proposal. There are a bunch of potential downsides that we can get into, but it's important to get set the facts straight about set the facts straight about where we actually stand today. So the core concept here is generally correct. Congress did pass a mandate directing the NHTSA to create a standard for advanced drunk driving prevention technology in new passenger vehicles. Now, the 2027 date that people are citing, that's not a hard deadline. The law was passed back in 2024, and in the actual law, it basically allowed for delays. So They said the NHTSA will only issue a binding mandate requiring automakers to actually roll out this tech when the tech is ready. And the NHTSA currently says that the technology is not ready. So in theory, the tech feels close, but the scale of the problem is so big, you can see why there's a delay here. Alcohol detection systems exist and are typically deployed for drivers with DUIs. You're familiar with blowing in the tube, probably, hopefully not personally. But the problem is that those little hoses that you blow into, those are active systems. They require you to actually sit there and do that for a minute. The government doesn't want that. That's not what the proposal is.
They want passive.
They want passive, which means breath sensing. So there's just like smell-o-vision, basically. If it smells alcohol, it doesn't turn on the car. The other one is fingerprint reading. So you put your finger on the start button and it scans into your finger and sees how many alcohol particles are inside. I guess that sounds sci-fi, but we're close, I guess. There's a couple other ways you can solve that. And then camera system. Just look at the driver. If they look drunk, then don't turn on the car. So all of these seem like they could be close to being roughly accurate right now. Like you can imagine an AI startup or a university lab putting something together at a hackathon that's 90% of the way there, maybe 99% of the way there, maybe even 99.9% of the way there. But the problem is that Americans drive a lot. The rough estimate is that there are almost a quarter trillion driving trips per year in America. It's basically every American, all 350 million of them basically, taking an average of 2 trips every single day of the year. So 224 billion trips a year is what the rough back of the envelope I did was. And what that means in practice is that if this system is 99.9% accurate, you're still looking at tens of millions of incorrect results every year. And the fact that probably 99.9% of these trips are not inebriated, like drunk driving is not 50% of trips, It's not 1% of trips. It's a very small amount.
I think about how people would abuse this new system. It'd be like students being like, "Sorry, I couldn't make it to the exam because—" Exactly.
"Dog ate my lunch." "My car wouldn't start because of a false positive on this." And so even if you're at 99%, you're still looking at tens of millions of incorrect results. The vast majority of those are gonna be people who are sober. Somebody wants to get in their car for their morning commute. They're a little sleepy, or they wore some cologne that triggered some sensor. System flags them as intoxicated and prevents them from starting their car, and it's infuriating. The tech will probably get there with enough time, time and effort. So it's worth looking into who supports this and opposes it. The mandate was actually bipartisan, but there's starting to be a backlash from libertarian conservatives who are worried about Orwellian government controls. There's an idea that there will be a remote kill switch, which leads to a bunch of dystopian possibilities. That is not in the current provision. That's not what's actually being proposed right now. But, you know, it's possible that at the end— that the end result of this process of back and forth, you do wind up with that exact capability. So people are worried about the system going off while you're driving at speed on the roads and then the car just shuts off and you get in a crash and that's like actually more dangerous than potentially the alternative. And so the middle ground seems to be what's called pre-drive lockout.
We need a tinfoil expert. The tinfoil enthusiasts have been saying that the sort of remote—
Are you using tinfoil enthusiasts to mean conspiracy theorists? I've never heard that before. It's good though.
That's a new one. Okay, yes. So tinfoil enthusiasts have been claiming that the remote shutoff button has existed forever.
Oh yeah, since the 1985 Bronco.
So when some sort of like, you know, witness or something like that just gets in a very inconveniently timed high-speed wreck, that is— Oh, okay.
I was unfamiliar with this conspiracy theory. It's interesting though. But yeah, so I mean, the current like consensus is around maybe pre-drive lockout being the more moderate solution than actually shutting the car off once it's driving. Still incredibly, uh, incredibly inconvenient if there's false positives. And then you also do still run into some potential negative outcomes where you go to the beach, you have a couple glasses of wine, you're not planning on leaving, but then the tsunami warning goes off, you need to get back in your car, and your car says no. Like, I don't care that there's a tsunami. Oh, I don't care that there's a tsunami. You have had 2 glasses of wine, you're at 0.08, you can't drive right now. and you need to tell the car, well, in this case, I'm okay with driving drunk because the tsunami's coming and I'm at the beach and I had a couple glasses of wine. And the car won't be able to potentially deal with that nuance, right? Right. And there's a whole bunch of other scenarios that could potentially play out. Just judiciousness is difficult, but potentially unlocked, AI agents, I don't know. You ask these models, what would they do in some certain scenario? Maybe there's a solution.
Yeah, I think there's a much, much stronger argument for rolling this out as soon the average vehicle has full self-driving capability. But then you don't need it because you can be as drunk as you want. No, I still think there's going to be this big window where you're not going to be able to— So maybe— specifically because if you need— as long as somebody needs to sit in the front seat of their car to potentially take over in any type of situation.
I like the idea of level 4 self-driving. You get in the car, there's one button you push to say, hey, just put it in self-driving mode. I've had a couple of drinks. Don't let me take the wheel. No matter how hard I try to negotiate. No matter how hard I try. It's all disabled. It's just Playmobil level. But then there's a second button. And for that, you have to do a full blood transfusion. They take— they centrifuge your blood and make sure that you have the purest blood possible to take the wheel and be able to actually drive the vehicle.
I don't know. There's maybe something there.
I did have a take on this, though, which is that Doug Gumerow, friend of the show, founder of carsandbids.com, has always talked about the eras of cars, the air-cooled Porsches. Then you have the fully manual, no electronic systems, no stability control, no traction control, those types of cars. Then you get into the manual gearboxes, the no turbos, no hybrid systems. The modern supercar era, it's just a bonus if it's not a full electric car. It's like, you know, it's going to be a hybrid, but is it, is it, is it, at least there's an engine of some sort. And that's what car collectors are sort of marking these moments where it's like, it's the last manual, it's the last non-hybrid supercar made by Ferrari or something like that. And there's going to be a world where you're like, oh, that was the last one that didn't have the camera that looked at you 24/7 or whatever. And I think it creates this like new class of like vintage cars. Investment. I don't know, maybe not investment grade, but certainly something that people— I mean, you can already see people reacting to this even though the post is like a little bit overhyped. You can already see people reacting to it being like, I got to buy a 2026 and hold on to it forever and keep it in great condition because I don't want that. But anyway, there's been a pushback on this stuff for a long time. Tyler, do you have anything on cars? Are you old enough to drive?
I was just thinking like, what do you think the premium is for a car that you can still like drunk drive in?
I think your car is— is gonna go through the roof.
This is not even worth joking about. Yeah. Anyway. Bad joke. Stop the fucking charade! Move on.
Moving on. What else is going on? We gotta talk about the GPT-5.5 prompt for Codex, which seems to have a duplicated line trying to get it to not talk about creatures. Tyler, you dug into this. What is actually going on?
Yeah, so it just seems like there was some emergent property of the new model where it always tries to talk about like kind of creatures and goblins, raccoons, ogres, trolls, ogres, pigeons. And so they had to counteract this, right? You have to put in the system prompt, tell it not to do this.
Where is this? Creatures. Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query. This reminds me of those old image prompts where like there would be a negative prompt that specifically said like, do not not put 6 fingers on the human's hand. But what a weird one and what an odd line to throw into a coding agent. Like, do we know anything else about this? Have people dug into like what is actually going on here?
I mean, it might not be just the coding agent. It's probably just the model in general. It's the model itself. It's Goblin Mode. The model itself yearns to discuss creatures, goblins, raccoons. It is an emergent property of superintelligence.
Uh, Vi says, they had to put this in due to my effect on the company. Goblins, creatures sort of followed me in through the front door when I joined, and we are only just now starting to understand the downstream effects of their presence. And Tebow says, never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it's absolutely— there's so many.
If you know, you know. Somebody says, my My 5.5 codec said goblin with a flashlight when referring to a bug fix yesterday. Star Wars meets Pawn Stars.
Have you seen this video? Have you watched this, Jordi Hays? I did. Yes. I saw the first second. I haven't watched the full thing, but I think it's pretty incredible.
I have items to pawn. All right, let's see what you got. Okay, so these are lightsabers? Correct. So, uh, where'd you get them? Estate sale. Estate sale. It's so funny that you can clock it by the audio more than the video.
It's very weird. Collector for many years. It is a great passion of mine. Each lightsaber has a story, a history. There is a certain flavor. I guess the character is CGI to begin with. So it's easy, it's harder to clock, but even he's probably fully AI in this video, or maybe it's edited together, I don't know. But yeah, you're 100% right, it's the audio that's awful. That's rough, man. How much are you looking to get for him? 100,000 credits. Look, I'll give you 50 bucks for him. You fool! It's interesting that, uh, why that's popular is because it's this mashup between two intellectual pieces of property and they're not being compensated for that. And so there's like this thing that can only exist in the piracy world, basically. Like if someone— if you had recreated this without leveraging Star Wars IP or Pawn Stars IP, doesn't go viral, right? And so the business case for AI video is still a little bit more narrow, and I think it will be tucked in the tool section. I was, I was, yeah, looking a lot at like the, the Aflac project that he sold to Netflix and some of the background replacement, some of the VFX stuff. Like it feels like the way, even though this is like this, oh, it's like one-shotting entertainment, it feels like the next moment of like AI in Hollywood is very much like tool-driven. Leveraging things like green screen removal and object replacement, VFX workflows, and just sort of speeding up remote, like rote tasks. But people will continue to have fun with these mashups, and I'm sure we'll see many more of them on the timeline.
The other news, Mike Isaac and his team over at the New York Times are in the courtroom live blogging. They're blogging. They went. They're blogging. Yeah. They're blogging. There's some crazy moments already, right? There's some crazy, crazy moments. Moments already. The judge is like, we'll try to put some of them together. The judge is doing bits. Apparently, uh, Elon's lawyer's microphone turned off 4 times in the course of his opening statement.
The screen turned off at one point.
I saw some different reporting on that. Apparently there were technical issues. 5th or so time, the judge says, what can I tell you, we are funded by the federal government. So that's why she's running— she's running bits. But highly recommend going over and— Yeah, Mike Isaac is fantastic coverage.
And looking at—
their posts are still rolling in. It's a lot to go through because it's hours and hours and hours of content. But we'll try to pull out some of the highlights and go through it on tomorrow's show.
The other thing you should listen to is Patrick O'Shaughnessy had Paul Tudor Jones, one of the greatest macro traders of all time. On Invest Like the Best. It's an hour and 11 minutes, and you should go listen to it. So go take a listen. It already has 5,000 likes, 1.7 million views, not nearly enough downloads. So go add it to your podcast player right now. Leave us 5 stars on Apple Podcasts, Spotify, sign up for our newsletter, tbpn.com.
Can't wait to see you tomorrow. See you tomorrow. Have the best afternoon of your life.
We love you. Goodbye. Flashbang. Flashbang. Out. Goodbye!
More from Jordi


SELL EVERYTHING (Including Your Dollars), Tesla Roadster Hype, Valar Atomics VC Feud | Reed Albergotti, Bryan Cantrill, Daniel Kimminau


OpenAI Bugging Incident, Are Your Devices Listening? Xbox Launches Remix Lab | Diet TBPN


OpenAI Uncovers Spy Device, Apple x LG, Home Devices Tier List | Thomas Neckel, Kabir Jain, Danielle Axman, John Urschel, Nicolas Sharp, Jason Ballard, Erik Bernhardsson, Josh Tyrangiel, Aida Baradari


Muse Wonder, Social Reckoning Review, Apple HomeView | Eli Manning & Michael Waldron, Anjney Midha, William Chu & Negin Oliver, Neal Stephenson, Ravi Gupta & John Hegeman, Jack Goldburg, Oege de Moor, Nick Talken, Jeff Holden


Crypto Going Bunker Mode, AI Mathpocalypse, Starbucks Explores Chipotle Takeover | Diet TBPN


Crypto Going "Bunker Mode", The Mathpocalypse, Starbucks Explores Chipotle Takeover | Rebecca Kaden & Michael Mignano, Nathan Benaich, Michael Sindicich & Healey Cypher, Moritz Stephan, Anastasios Angelopoulos, Zach Yadegari


OpenAI Math Breakthroughs, SpaceX’s $40B Nvidia Deal, Quant Firms Hiring Idea Guys | Diet TBPN


Some Math Problems, Arsenal-2, Porsche Mission S | Walter Russell Mead, Kat Cole, Tarek Alaruri, Tyler Niday, Scott Morton, Josiah Meyer, Aarush Selvan


AI’s Economic Impact, Warren Buffett Discovers YouTube, Mistral Launches “Le Chonk” | Diet TBPN

