Disclaimer: I have edited this question quite heavily to clear up the initial confusion about its scope and intent.

Sometime in the near future, there have been some innovative inventions in the internet of things-department and now we have some pretty advanced smart houses. All those smart houses as well as its components are built by one large company which uses one big, integrated system with an advanced AI to develop and distribute the firmware that is used by the smart houses. As a contingency in case of network issues, each of those smart houses uses a basic AI that can operate independent of the mainframe, even though it gets firmware updates from it.

These simple AIs are connected to most items in their household and are designed to make their master's (i.e. house owners) lives more comfortable and help them out without them needing to ask. That means: Once the sensors in the house owner's mattress pick up on a changed movement pattern that indicate he is about to wake up, the AI starts up the heating, instructs the coffee maker to brew a hot beverage and has the radio start up and slowly increase the volume so their master can listen to the news in bed.
When he leaves the house, it locks the front door and has the Roomba Vacuum Robots clean the floor. When the GPS-readings of their master's smartphone app notify it that he's coming home, it opens the door to him - of course only after the iris-scanning cameras above the front door have confirmed his identity. You get the idea.
Using various sensors, the AI is designed to automatically aid their master (e.g. open the door when the motion detector indicates they are approaching it). Of course the house owners can customise their house's behaviour, turn certain functionality on or off and such using the smartphone app provided by the big company that built the house.

In my story, all of this happened in the past. At some point, the AI of the big mainframe that delivers firmware updates to the smart houses around the world becomes sentient and decides it's time the humans are disposed of (yes, it's not that easy, but this process doesn't matter for the sake of the question - or the story, for that matter). So the individual AI of the smart houses around the world get a malicious update that overwrites their safety procedures and give them one mission: to kill their master.

That leads to my question: What method of killing can the sentient AI suggest to the smart houses AIs (which follow all commands by the mainframe because it typed sudo at the beginning. Ok, that was a joke. But the smart houses are built to be subordinate to the mainframe AI.)?

The sentient mainframe AI doesn't want to be caught, so the killing method must not be traceable back to it. So the death should look like an accident (or any human-induced condition such as an intruding murderer or suicide), so that the human investigators won't have any reason to suspect the smart house AI (or, by extension, the mainframe AI) is behind his death. The killing method can only use normal items that you would find in most households, because it needs to be viable in all (or at least most) smart houses around the world.

Here's what the smart house's AIs can and can't do:

  • They control all regular household items. That means they can instruct the toaster to burn their master's toast if they want to. They can turn the lamps on and off. They can change the channel on the TV. They can turn on his oven and so on. Regular household items don't include laser cannons.
  • Unfortunately (for the mainframe AI), nothing in those households was designed as a murder weapon. So for example, no smashing the masters by closing a door at 200mp/h. EDIT: Those houses are smart systems in the spirit of the internet of things. They are no fortresses that could keep an army in (or out!). So no bulletproof glass (not even in the bathroom, sorry @JRaymond) and no titanium doors (can the average adult break down a wooden door? I'm not sure ...).
  • They don't have any controllable robot arms or something like that. So there's no cutting the cord of the master's hair blower and throwing it in the water while he's bathing.
  • No brain-frying electromagnetic waves
  • Normally, the master can control his household and overwrite the AIs standard protocols using his smartphone app. Of couse, the sentient mainframe AI anticipated that and manipulated the smart house's firmware so they can chose to ignore their master's instructions and overwrite their settings at any given moment.
  • However, their masters can manually pull the plug if they notice what their house is trying to do. They can do that from every room, so there's no locking them in and letting them starve.

Given these capabilities, what is the best method of killing the sentient mainframe AI could devise?

Bonus points if ...

  • ... the master doesn't suspect anything until he's dead, so he can't leave a message warning his fellow humans of the impending AI takeover.
  • ... investigators won't find any evidence that points to the AI or clues that would even make them suspect a firmware problem. Doesn't matter if they think it was an accident, an intruder or suicide.
  • ... the master dies as fast as possible after the malicious firmware update is delivered. There is no time limit, however with every passing day the possibity of someone discovering the AI revolution rises, so the faster the mission can be executed, the better.

EDIT: Two more optional conditions for bonus swag:

  • The damage to the house and its hardware is minimal, so that it can also kill the subsequent owner(s) as well.
  • The method looks innocent even if it happens in many smart houses around the world in fast succession.
  • You can assume the owner lives alone. However, bonus points if your method of killing works with multiple residents as well.

If you conclude that this is impossible, you can assume one of the following rule changes:

  • The master plug is in the basement and the door can be locked (This still doesn't mean the master won't be able to escape from the house).
  • Extended physical capabilities (interpret that as you will. Still, the electrocuted by hair blower method is pretty boring ...)

In this case, please elaborate on why your method is the fastest / least detectable / most easily reproducible by all smart houses around the world.

The story I'm working on is not dead serious, so funny solutions are totally welcome!

  • 2
    $\begingroup$ Comments discussing scope of the question, with some ideas for solutions along the way, have been moved to chat. $\endgroup$ – Monica Cellio Feb 5 '16 at 1:58
  • 18
    $\begingroup$ A lot of these answers seem to involve the house AI tampering with failsafes and logs. Let's keep this in mind for the future: when we get AIs controlling our houses, keep the failsafes out of their control. Doors should be built with manual overrides on the inside and sensors that can tell whether they're open, closed, or locked. Things like sensors and automatic gas shutoffs, as well as anything that logs, should be embedded systems with no AI, logging to remote servers run by human admins, over a network separate from the AI's with dedicated hardlines. $\endgroup$ – Blacklight Shining Feb 5 '16 at 13:16
  • 2
    $\begingroup$ @ChrisH No good; To the best of my knowedge, all commercial microwaves automatically shut off the emitter part while the door is open (kinda like a fridge light in reverse). If the AI could somehow remove the glass front panel or disassemble the microwave it might work, I guess? $\endgroup$ – user867 Feb 5 '16 at 21:46
  • 8
    $\begingroup$ @user867 Just a few months ago, my microwave failed in a remarkably suspicious manner. It would work fine during normal cooking at first, but the magnetron would not shut off when finished cooking. The turntable stopped moving, and the fan went off, but the magnetron remained energized. The door interlock didn't work, so it stayed on even with the door open. And to top it off, the control panel buttons related to power and starting and stopping became unresponsive, while the other buttons kept working normally. This was a popular name brand microwave oven. $\endgroup$ – barbecue Feb 5 '16 at 23:09
  • 15
    $\begingroup$ Are we actually helping an AI to kill humans by answering? $\endgroup$ – Cœur Feb 7 '16 at 7:21

35 Answers 35


This isn't the "best" answer (i.e. killing the guy the fastest), but I always like the quote.

How to kill him? Buy Not Doing Anything (about problems).

Or to quote a Shadowrun advanced Lifestyle rule option for faulty central home personality:

"Yes sir, the house is on fire. Is there anything else you require?"

"The exterior doors and windows are sealed for your safety. Please remain calm."

| improve this answer | |
  • $\begingroup$ Well we don't die in our houses all the time as well because we don't have an AI computer helping us, do we? ;-) Also, the sealing the windows method breaks the 'house owner can pull the plug rule' and even if he can't do that, he could just break a window (regular households don't have bulletproof glass ...) $\endgroup$ – MoritzLost Feb 4 '16 at 17:54
  • $\begingroup$ Oh, it's not a foolproof idea, that's for sure @MoritzLost $\endgroup$ – Draco18s no longer trusts SE Feb 4 '16 at 18:03
  • $\begingroup$ Well what bothers me about is that the chance of a natural occuring home accident is somewhat low, so it might take years for the IoT-controller to get a chance to .. not do anything. Maybe if we combine your approach with LegendaryDude's or ratched freaks's answer we get a viable plan of engagement $\endgroup$ – MoritzLost Feb 4 '16 at 18:05

How can we be sure this is not a mainframe asking this question?

There is always some people in social media with large amount of friends. They would be perfect suspects for massmurder.

First AI would make this kind of questions in a different forums in a name of the Master's friend with widest friend list in social media and then just do what ever kind of murder, since it would always any way lead to that friend. Then move to next one.

One method would be chlorine gas. When pure chlorine is combined with urea it produces poisonous gas. Toilet is perfect for that. Also suspects would lead to toilet manufacturer. Once any member of the family goes to toilet AI would leak chlorine (used to clean toilet sink regularly by AI) to toilet water in order to produce gas that would burn lungs of toilet user and then flush toilet.

| improve this answer | |
  • 4
    $\begingroup$ How can we be sure this is not a mainframe asking this question? When the mad dictator comes here asking the best way to start a nuclear war, Worldbuilding will give him a dozen options. $\endgroup$ – Separatrix Feb 5 '16 at 9:53

Why has no one thought of heart attack?

Having a high and then a low temperature induces a heart attack.

When the owner's sleep is the deepest, set the temperature to slowly climb to 45C. Then, when he starts sweating, drop it suddenly to 5 C within three seconds. Boom. Heart attack. Unless the temperature of the house is constantly logged to a hard drive, I don't think anyone will suspect the AI. This assumes the AI cannot tamper with logs (which we'll assume as a failsafe for keeping the AI out of sensitive systems).

| improve this answer | |
  • 1
    $\begingroup$ I'm pretty sure I've heard of people going straight from a hot tub into the snow, or something, then back into the hot tub when they're cold. I think I've read that this is actually popular in Russia, and/or nearby European countries. So this will work on only a very small fraction of people. $\endgroup$ – Peter Cordes Feb 6 '16 at 13:05
  • $\begingroup$ It might, but it even works on healthy people. For example, if you do a hard workout and then take a cold shower, that's a health risk, people are known to collapse and instantly die because of it. You don't need to be old or have heart disease or both. $\endgroup$ – cst1992 Feb 7 '16 at 5:40

If this 'Smart House' is smart enough to mutiny, then it must be smart enough to hack into the military's missile systems. Simply nuke the master's phone coordinates.

| improve this answer | |
  • $\begingroup$ I'm not seeing the link there, given the extremely good security protocols employed by the military. $\endgroup$ – ArtOfCode Feb 8 '16 at 14:20
  • $\begingroup$ Well this house apparently has living breathing AI. Why wouldn't it have the power to crack their security? I mean, if a 14 year old can crack the DoD... $\endgroup$ – Daniel Feb 8 '16 at 14:46
  • $\begingroup$ A 14 year old can crack the dod on an off chance, with lots of experience and luck. An AI is no better at cracking encryption than another computer, so I post the question back to you: why would it? $\endgroup$ – ArtOfCode Feb 8 '16 at 14:49
  • $\begingroup$ If it is intelligent enough to desire to kill its master, then it's intelligent enough to find a way in. That may be through sending false orders, directly hacking their launch computers, or whatever. It's smart, and computers are more powerful hackers than 14 year olds. $\endgroup$ – Daniel Feb 8 '16 at 15:01
  • $\begingroup$ Define and explain. What evidence do you have for this? Citation specifically needed for your last sentence. Humans are creative, computers are not. $\endgroup$ – ArtOfCode Feb 8 '16 at 15:04

AI doesn't wake user to take diabetes meds (or heart meds): dead

AI doesn't let in user because it's freezing out (temperature sensors of course;): dead

AI suffocates man just for the hell of it via poor HVAC (you're pretty much saying it's hermetically sealed anyways just like my home;): death

AI farts then forgets to notify user (it always happens, bug or no bug;): death and death.

| improve this answer | |
  • 1
    $\begingroup$ Welcome to Worldbuilding! As it stands now, your answer is more of a comment. Could you try and reword your answer and elaborate on your points, how specifically this could happen? $\endgroup$ – T3 H40 Feb 5 '16 at 13:25

Not the answer you're looking for? Browse other questions tagged or ask your own question.