NASA, AI and One Rogue K2 - That Time my AI Called the Police

Rogue One is one of my favorite Star Wars movies - Excellent acting, a great story that fits neatly into the bigger overall story, and - arguably I’m sure - the coolest Droid in the cannon - K2. This is the story of how my K2 went seriously Rogue to the point that the police were notified.

I’m going to start on a bit of a tangent, and then get to the point. I started down a massive rabbit hole about a year ago. I was on a call with an engineer and he mentioned a device called a Hubitat. The Hubitat can interface and control smart devices in your home, effectively a completely local - and more powerful - Google Home or Alexa. I got one to tinker with, and quickly found that it was pretty clunky. Some cool integrations, but I didn’t love that configurations are not easily accessible as yaml, etc…

That led me to Home Assistant. Which I love. It is absolutely an excellent way to rabbithole. The number of integrations and power of the platform is crazy. Love it. So, I got that, set it up. and got started.

One of my first projects was a lamp (and eventual pixel clock) that tracks my blood sugar. Pretty cool! If my blood sugar goes over 150, the lamp turns red. It goes blue if I drop below 60. As a type 2, I found this to be awesome to remind me to get up, do a few pull-ups or go on a walk when my sugar spiked. BUT, I digress. The point here is I started building things again. Tinkering. So, of course, I needed a 3D printer (of course). So I got one…of course. It’s a Bambu labs X1 Carbon. I highly recommend picking one up…if you have the means. The A1 is much cheaper and also truly an excellent 3D printer.

We are getting to the point, I promise.

So, several projects in - of course a space camera, a high resolution smart bird feeder with some bonkers AI and ML, and…my K2. I’m not very good at traditional crafts, so I have to say that I impressed myself with K2. I printed him, sanded him, gave him a metallic undercoat, and tried to match his scrapes. Turned out!

With his voice and light up eyes, it’s pretty cool, fast, responsive, and with the exact same snark he had in the movies. He even thinks that my house is a stolen Star Destroyer called “The Loop” and that my dog (Bast) is an alien. He might be right actually.

Next of course was his voice and AI. I used Elevenlabs to clone K2’s voice from the movie. It really really works well. And then hooked him into HomeAssistant to set up the AI with OpenAI’s conversations. There’s an option to give the Assistant access to devices - similar to Google Home’s voice features. So, of course, I checked that box (what could go wrong?)

A little time passes. I added a new feature to my Space Camera, which is an ADS-B received. Basically it reads the hexcodes from planes so that when a plane is in view of my camera, it will say the tail sign. Cool! So then, of course, I decided to enhance that. I used some things like flightaware to get more information: who owns the plane, what kind of plane it is, where is it headed or coming from, and how fast is it going. The space camera itself is also…awesome. This is the Flame and Horsehead nebulae:

NEXT, I had a FANTASTIC idea. Why not have K2 rate the planes? GREAT IDEA. Rate it on a scale of 1-10 based on the owner, speed, altitude, destination, etc… and then give your typical snarky summary of the cool ones. Make up a story if need be. It makes for a lot of laughs. So then - of course - that lead to another idea:

It should alert my phone if there’s a 8 or above. Send me a message! Of course. BUT THEN….

Something funny happens. I come home and the lights are on in my bedroom. I didn’t turn them on. Weird. I kinda think nothing of it until later. It happened again. I looked at the Home Assistant logs, and - my my surprise, K2 had turned them on. IT decided that the plane flying overhead was intesting enough that maybe it should alert my phone AND turn on my bedside lamp. Kinda wild - and interesting! Maybe time to rethink access??

BUT THEN…I was at work. And my house alarm goes off. Full Blast. Wont stop. THE ALARM SYSTEM CALLS THE POLICE. VERY F’ing weird! I was finally able to shut it off remotely. AND guess what I found in the logs. Yep. K2 saw a NASA jet flying at 40k feet over the house and thought, THIS IS MY TIME! 10 OUT OF 10.

This is the exact code of the automation. It VERY explicitly gives the action, and nothing else. Notify.mobile_app_pixel_9_pro_nick. which is a command to send a message to my phone:

      - data:
          message: >-
            🛩 Interest Level {{ rating }}/10: {{
            response.response.speech.plain.speech[9:] }}
          title: Aircraft Alert
          data:
            photo: "{{ photo_url if photo_url else '' }}"
        action: notify.mobile_app_pixel_9_pro_nick

SO. It decided, you know what, “A text message JUST WON’T FLY here. That is not what I’m going to do. I’m going to, instead, SET OFF THE HOUSE ALARM”.

So. I take away house alarm access, obviously. BUT - What this made me think about - is We do not know enough and We cannot predict the unintended consequences of what we’re doing!

Sure it’s fine and funny for this situation, but what about this? https://euroweeklynews.com/2025/01/16/tesla-will-produce-army-of-robots-called-optimus-by-2027/

We may think we’ve told these robots exactly what to do. But have we? I think we’ll find there’s a lot we haven’t thought about. We will NOT think around all of the corners.

Anyone have similar stories?