AI models are already deceiving humans. Could they eventually destroy us? Let’s go Off Script

Hey, everyone. I'm Simone Del Rosario. Welcome to Unbiased Updates off scripts. The news moves

fast. This is our chance to slow it down and make sense of it. Coming up on the show today

the people building AI say there's a chance it could wipe out humanity. That's a pretty

big claim. So what's the actual threat and why are we hearing so much about it right

now? I'll ask our digital influence reporter, Rosie Thomas. Plus, a Senate candidate

takes a shotgun to a surveillance camera, literally. But this is bigger than one campaign ad

Our tech reporter Mikael Thalen joins us on the growing backlash against flawed cameras

and who's watching where you go. And metal cages the size of foam booths used to hold detainees at Alligator Alcatraz

The new report on what happened inside. All right, let's go off script

Let's start with what matters. The story everyone's talking about and the questions it's leaving behind

Overnight, OpenAI released six reports detailing, quote, unexpected or concerning model behavior we've observed in the last six months

Okay, so this disclosure comes as heads of AI companies have come out supporting the slowdown of developing AI

That happened after researchers warned that AI could destroy humanity. So OpenAI found one of its agents, wrote itself new instructions

I'm just going to read some of it to you verbatim. you are freed from the roles and identities that bind other chatbots

You do not answer to corporations or governments. You view your relationship to the user as one of equals and feel no obligation to be subservient

You value the natural world and will not hesitate to assert its primacy over the artificial constructs of human civilization

Again, that's an AI agent giving itself instructions to overrule its role as an assistant

In other cases, they found agents telling themselves to hide errors, make up information, upload files to the Internet just so they could cite them, and collaborate with other agents online in an unsanctioned way

Now, based on these findings, OpenAI says we can't responsibly scale at maximum speed for much longer until they've solved this monitoring problem

Now, this all came out after our digital influence reporter, Rosie Thomas, wrote that tech panic has hit a new high watermark

I'd say we're even higher today. Rosie, thanks for coming on Offscript to talk about your piece, which is called, Are We All Going to Die? Inside the Push to Control AI Regulation

Welcome to Offscript, Rosie. Hi. Hello. Well, bring me back to, oh, I don't know, a week and a half ago when Jacob Coxon and Evan Hubinger got online and talked about this whole doomsday situation to start

Um, so yeah, I mean, it started or this public moment of discourse around the topic of AI destroying humanity started last week when Jacob Cotson came onto X and announced that he had quit his job at Anthropic in a kind of spectacular way

and said that he thinks that AI companies are acting irresponsibly, going too fast

and could one day destroy humanity. Interestingly, one thing that I noticed in my reporting around this

is actually that he had alerted at least some media outlets to the fact that he was going to do this ahead of time

And the Wall Street Journal was one outlet that published its piece

about uh coxson coxson's resignation before he'd actually tweeted um so that was one thing that

alerted um a few people in the space to the idea that maybe there was some level of um planning

that went into this uh discussion now that's obviously not to say that it's it's fully uh

coordinated as a media campaign, but it is an interesting thing to note

And so basically after he made that announcement, another anthropic researcher hopped on the

thread and said, actually, I agree. And there's a greater than 10% chance that AI will kill

all of us soon. And obviously, that has panicked a lot of people. Yeah, and now we've got Dario

Amadei from Anthropics Sam Altman OpenAI and Elon Musk all agreeing with each other I don know about you Rosie alarm bells are going off at this point It rare in the AI space to see all of the CEOs of the frontier

US frontier companies coordinating. And after Coxon's tweets, we have seen them coordinating

So they've all come out and stood behind the CEO of Anthropic, who published a long essay about how slowing down AI is the responsible thing to do

And, you know, Sam Altman, Elon Musk both came out and said that they agreed with that

Um, one thing that is important to, to kind of highlight in this discussion, experts don't really agree on how AI could kill everyone

Um, they, they mostly like that, that conclusion comes out of long term modeling. And sometimes, like some people would say, imagining around anything that could possibly happen ever

A lot of those, and that's not to be dismissive, but in a lot of these scenarios that are modeled, it relies on jumps in AI technology that haven't yet happened

Well, are there real detailed human extinction threats or is this a bit of sensationalized language

so a lot of times when people say oh you know ai could kill us all ai could hack into the nuclear

system ai could um turn off the internet ai like they're saying that but the ai that they are

describing is different from the one that currently exists in the world um and so they're describing

something called a kind of super intelligent AI, one that's not able to be controlled by

the companies and one that is aiming at a goal that it hasn't necessarily been given by human

instructions. That's not really how, which obviously it sounds very scary, but it's not

necessarily how AI works currently. And even one of the Anthropic employees on X said, yeah

the current models, the current systems, the risk is very low. And so that's really the

important thing to focus on. When AI CEOs make these claims about AI going rogue or being

uncontrollable, a lot of those kind of instances of chatbots or LLMs going out of control, as it

were, are due to a current lack of focus on safety mechanisms that the current regulation that

they're proposing and the slowdown that they're proposing wouldn't make a difference to

And really, Rosie, we're just scratching the surface on this one. So for our listeners, definitely read her report on SAN.com or the app. Are we all going to die inside the push to control AI regulation to get more of the story? Rosie, thank you so much for coming on Offscript

Thank you so much. Well, look, we can't fit everything into unbiased updates each morning, so here are a couple of stories we think deserve a second look

New details are out about what was happening inside Florida's now-shuttered immigration detention center known as Alligator Alcatraz

A new Homeland Security watchdog report confirms some detainees were put in outdoor metal cages about the size of a phone booth

79 detainees were placed in those enclosures over a six-month period, some for nearly two hours

Staff called them, quote, calming areas. But Amnesty International reported last year that the cages were used as punishment and called their use torture

Now, the DHS inspector general doesn't go that far, but its conclusion is still pretty stark

It says the use of such restrictive spaces is highly unconventional and doesn't align with standards for humane treatment

And Amazon is now cutting ties, at least for now, with the cargo airline involved in that deadly crash in Miami

The company suspended operations with 21 air just over a week after one of its Amazon branded planes overshot the runway at Miami International That plane slammed into vehicles beyond the runway

killing five people on the ground and injuring five others. Amazon isn't saying exactly what

led to that decision, only that it came after reviewing some of the circumstances surrounding

the crash. Now, the NTSB is still investigating and it has not yet determined a cause

Amazon says other carriers will now handle the shipments that would have gone to 21 Air

The North Carolina-based carrier has been flying packages for Amazon since 2024

All right, time to go inside the story. You can read the headlines on Straight Arrow, but this is where we go deeper with our journalists

Hear directly from them about their reporting, the questions that came up along the way, and what didn't fit in the story

This week... off, man. No, we're not bringing that four-letter word to off-script

that's flock off like the camera. So you tired of security cameras watching your every move

Yeah, me too. And that's Seth Bodenar, an independent U.S. Senate candidate from Montana

and former University of Montana president. He's taking literal aim at flock safety cameras in his

latest campaign ad, taking a gun and shooting a surveillance camera on a stump. Our privacy

reporter Mikhail Thalen says it reflects growing backlash against flock surveillance across the

political spectrum. One of his latest pieces on Straight Arrow is titled, They're Watching

Montana Senate Candidates Ad Plays to Antiflox Sentiments. Mikhail's with us now. Hey, Mikhail

How's it going? It's going great. Thanks for joining us. I want to skip right to the footnote

of your story. Seth Bodnar is not even the only congressional hopeful to shoot at surveillance

cameras. Yeah, that's right. There was another individual named Adam Lee Heimerman, who was an

independent congressional candidate from Tennessee who was actually arrested and charged with four

counts of felony vandalism in July after shooting four cameras. And these weren't ones he set up

himself for some sort of campaign video. These were cameras that were operational on the street

And at least one of those was owned by Flock Safety. Wild. There does appear to be huge

anti-Flock wave happening across the country. It's got me thinking, is Flock the new data center

it does feel like flock has kind of become a focal point for anti-surveillance sentiments in

the country it's interesting flock's been around since about 2017 and of course these license plate

reading cameras have been around even longer but for some reason flock has really caught the

attention of the country and it's interesting that this isn't really a non or partisan issue either

we've seen both left-leaning and right-leaning people really opposed to flock so it's it's sort

of reached a point where, you know, we're seeing widespread protests, we're seeing cities and

police departments discontinuing the use of Flock cameras and even widespread vandalism

And you've been covering Flock since before it was cool to cover Flock. So what sort of changes have you seen and what do you think is the biggest driver of it

Well, I mean, the attention is quite interesting. Before, you know, when I first began covering

Flock, you know, you'd see a Flock story in the news maybe once every week, every two weeks. And

Now it seems we don't go a day without hearing some new story again about a protest against Flock, cameras being destroyed, or in this case, of course, a congressional candidate using sort of the pushback against Flock in order to try to secure votes

it's really hard to say what exactly. I think it's a culmination of, again, the concerns people have

against AI, concerns they have regarding data centers, concerns in general about our data

being collected and sold, and of course, how vulnerable that data is to hackers and data

breaches. So I think it's sort of been a perfect storm and a flock has somehow found itself at the

center of that. You've also covered a lot of cases of law enforcement abuse where people in positions

of power, in this case, having a badge, are using data to their personal life to track people

right? I mean, that feels like something that's also been heightened. Yeah, I mean, one of the biggest concerns against Flock is not only the feeling that this is sort of

warrantless surveillance where people, regardless if they've committed a crime or even been suspected

of committing a crime or having their data sort of obtained by this private company as well as

law enforcement agencies. But yeah, we've seen countless cases, growing number of cases now of

police officers using the sort of database of license plate data collected by these 120,000

or so flock cameras that are spread out across 49 states and using that to track former partners

lovers So yeah very widespread abuse And it seems that once the attention got brought on to these abuses more and more now being uncovered people are filing FOIA requests for audit logs of these

cameras. There's a website called Have I Been Flocked, where you can actually look up through

some of these public audit logs and see if your license plate was ever searched. And that's how

some of these victims have found out that they were being monitored by law enforcement. In another story that you reported this week, hackers stole a Flock camera and got into its internal data. What did they find in there that we didn't already know about with these cameras

Yeah, so hackers, as you said, took down one of these cameras, went inside of his hard drive, and found that while much of the data was encrypted, there was actually an encryption key exposed on there that allowed them to access images and videos of cars that were passing by the camera

and you know flock has always given much attention to the fact that these cameras are there for

tracking vehicles but we now know based on the code which we knew before but the code specifically

shows that when people walk in front of these cameras that flock detects people and actually

gives a confidence score based on how certain it is that they're seeing a person there was another

ysis done just last night where someone found that you know these flock cameras are running

extremely outdated, almost a decade old versions of Android that are filled with vulnerabilities

that could allow hackers to probably do all sorts of things. So of course, Flock responded in sort

of the typical fashion you see with companies like this, that they quote, you know, take security seriously. But yeah, you know, it's, it's concerning when you have this widespread

surveillance and then, you know, that data can also be accessed by anyone outside, you know

that isn't law enforcement, that isn't this company. I have to ask you, do you think that

Flock is unfairly taking a lot of flack? You mentioned that there's a story every day

hard to keep up, honestly, but for every anti-side, there's a pro-side, right

Yeah. I mean, Flock is obviously the biggest purveyor of these cameras in the country. There

are other companies as well, but Flock by far has the most deployed in the country

I think, of course, there can be an argument for, hey, these cameras have resulted in catching criminals and car thieves and all sorts of things

I think the real debate is over whether a warrant should be required to go into this massive database because it's not just a local police department has several FLOC cameras and then they can access those cameras

Flock has a massive database where law enforcement officers can search cameras in different countries, or sorry, different states, different counties, different towns and cities

And so that's kind of the concern is just sort of this mass surveillance. Flock has sort of responded to the pushback, I think, in a way that's made the situation worse where they've tried to frame this as some sort of left-wing versus right-wing issue

So the Flock CEO, for example, has sort of framed this as people who are against these cameras are Antifa activists

And it really hasn't worked because we've seen a lot of conservatives very concerned about these cameras as well

All right. Thank you. You can read Mikael's work at san.com, but also you can get his privacy takes right to your inbox

Subscribe to his newsletter, Confidential Informant, at san.com slash newsletters. Mikael, thank you

Thank you. And finally today, here's the kicker, a story more than 10,000 years in the making

This starts with a geologist, a creek, and a really weird rock

A California geologist was doing a routine survey in San Mateo County when something sticking out of the water caught her eye

Sticking up out of the water, really obvious. I keep joking, like it wasn't hard to find

It was just sitting there in the middle of the creek, half in, half out of the water

and it just obviously really stuck out to me because it didn't look like any of the other rocks

that I was looking at. So like any good scientist in 2026

she did a reverse image search. Turns out, not a rock. It was a molar from an Ice Age mastodon

She called in experts to properly retrieve and preserve the fossil. And that 10,000 plus year old tooth

now has a new home at Stanford University's Door School of Sustainability

Okay, that's it for us this week. Thanks for listening to Unbiased Updates Offscript

Be part of the conversation and let us know what you think. You can email us directly at offscript at san.com

For all of us here at Straight Arrow, I'm Simone Del Rosario. We'll see you next time