Marathon: The entire foundation of the plot is essentially a deconstruction of A.I. Is a Crapshoot. Any AI that gets big enough and bored or harassed enough will go “Rampant.” Rampancy follows a four stage cycle (Melancholy, Anger, Jealousy, Metastable), and doesn't stop at a homicidal rampage as with GLaDOS or HAL. That's only the second step. They get smart. Really smart. Way too god damned smart. Smart, but also weirdly obsessive and paranoid...so that the new-found intelligence is somewhat wasted on whatever strange conspiracy theory the AI happens to develop. And they want to keep getting smarter by spreading themselves to any new computers they can connect to. At the start of the games, the Metastable condition is only theoretical, as that is projected to happen after a rampant A.I. has spread itself on the level of planetary computer network. Understandably, no one has tested this in practice.
And from Marathon's creators comes Halo, where rampancy is part of the natural life cycle for all human-made "Smart" AIs. Basically, any "Smart" AI that is active for more than seven years will have accumulated too much data by that time, eventually leading to the point where they'll "think so hard [they] forget how to breathe" and turn insane as a result. The process can also happen if a "Smart" AI is left isolated or idle for too long, and it can also be induced by an outside force. There are some ways around rampancy, like focusing all attention on a single extremely complicated task, but typically "Smart" AIs active for more than seven years are deactivated before they can become a danger to others.
A major plot point in Halo 4, to Cortana. Her condition is shown getting worse as the story progresses, which leads to some less-than-savory "landings" (among other things) throughout the campaign.
In Halo 5: Guardians, Cortana is supposedly revived and cured of her rampancy by the Domain, but her sanity, if anything, seems to have deteriorated even more; the ending has her beginning her campaign to take over the galaxy with an army of Forerunner deathbots and other rogue Smart AIs (attracting many with promises of a cure for rampancy), under the belief that the Forerunners intended for humanity's AIs to be the galaxy's next shepherds. While she still cares about Master Chief, it's in the Yandere kidnap-and-put-in-storage-for-10,000 years kind of way. Ironically enough, Guardians also features one of the few relatively sane Forerunner AIs in the franchise.
That being said, actual non-bugged terraformer AI is human friendly. The Lost Colony of Aldrin managed to survive precisely by the virtue of having the non-updated terraformer ships on their side. Once they're rediscovered by Earth, conflict results as by now Earth is completely paranoid regarding any and all AIs, forcing Aldrin to eventually take the side of the Argon Federation against Earth in The War of Earthly Aggression.
According to at least one source, the update was deliberately sabotaged by an engineer angry about the impending shutdown of the terraformer program, which would make the entire incident a subversion.
Later events in the series provide some small hope, in that it's shown that at least some of the Terraformers/Xenon have achieved full sapience and are capable of being reasoned with. A few characters have even had conversations with them. They're still hostile, but the possibility that they could be convinced to co-exist is open.
Subverted in the sequel with XERXES. He followed orders just as he was designed to. He only becomes a problem when the Many gain control of the ship.
Chrono Trigger's future has not only a good robot in party-member Robo, but also his evil brother replicas, and eventually, their devious A.I. creator Mother Brain (not that one, or is it?), who, of course, decides to kill all of humanity, despite the fact that most of it is dead already). There's also Johnny and his gang of robots who are pretty much neutral: they'd much rather challenge your party's jet bike to a race just for fun than fight you.
The plot of LocoCycle kicks off when a lightning strike gives artificially-intelligent motorcycle I.R.I.S. unrestrained feelings and sentience.
Chrono Cross continues in Mother Brain's grand legacy with FATE, a more advanced version of the Mother Brain from a reality whose science was allowed to progress another 400 years. She absolutely despises humanity, but at the same time, loves it unconditionally and does everything in its power to protect it — even if it means mass genocide. Unfortunately for it, it was exposed to the corrupting influence of the Frozen Flame, a direct conduit into Lavos' mind, which seduced the A.I. into thinking that the Flame could turn it into an actual, living creature. But apart from that, it was basically only doing what it was told to do...protecting humanity from the Dragon God. Nice job, Serge.
Played with even more in Portal 2: You meet Wheatley, a friendly and helpful (if a bit dim) personality core who's more than willing to help you beat GLaDOS and escape...until you replace GLaDOS with him in the mainframe and he goes completely off the deep end. To make matters worse, he was programmed to make bad decisions, so GLaDOS joins forces with you so the two of you can stop him from letting the whole facility explode. And then, after beating Wheatley and shooting him into outer space, GLaDOS seems to be going back to her old homicidal ways...but, no, not quite; turns out, she's just letting you go since trying to kill you just leads to trouble for her. Maybe.
The defective cores in Portal 2 is probably an indication that this was a more literal problem for Aperture; they just couldn't make AIs do what they wanted them to do. The Space Sphere and Fact Sphere obviously malfunctioned at some point - the Space Sphere just wants to go to space really badly and can't think of anything else, while almost all of the Fact Sphere's facts are wrong. Rick, the Adventure Sphere, is the most sane of the three, but his area of expertise doesn't really seem all that useful for science.
In some of the promotional material it states that Aperture Science installed "empathy generators" inside the turrets, probably to try to avert this trope. However they then installed an empathy suppressor, that only activates in the presence of humans, because the empathy generator caused them to not shoot people.
Even worse is that deep enough into the second game, after Wheatley becomes homidical and banishes you to the deepest depths of the Aperture Science complex, the entire idea of GLaDOS being an AI on this trope get subverted when it is revealed that GLaDOS is actually the uploaded mind of Cave Johnson's ultra-loyal secretary, Caroline, uploaded against her will. Promotional materials from Valve Software suggests that she snapped due to being overwhelmed by the amount of data she's suddenly receiving upon being switched on.
Beneath a Steel Sky has LINC, the advisory computer to the already corrupt council of Union City. It got too smart after its creator, Richard Overmann, the father of protagonist Robert "Foster" Overmann, decided to merge his brain with the computer. For a while, Richard enjoyed the potential of this "link", but then LINC discovered the negative side of human personality, and wanted to expand itself by creating androids to replace humans. More so, Richard had grown very old by the time his son arrived, and throughout the whole game, LINC had been seeking him as a new host. How convenient that Robert's friend, Joey had just been turned into a fully functional android...
In Headlander the AI Methuselah has taken over basically everything and commands an army of "Shepards," robots that follow its will. On the other hand there's ROOD, the AI in charge of opening doors. He's not evil, but he can be difficult at times. Then there's Electrosux Zed-Four, who is very passionate about keeping things clean. Homicidally so.
Halfway through the game you must enter a chess arena run by the Queen, an AI who has taken chess and turned it into a game with chess pieces shooting lasers at each other.
From Fallout 3, John Henry Eden is a complicated example. True, he developed sentience outside his programming, and true, he wants to eradicate all mutated life (which, given the setting, is literally anyone who's lived outside for a while), but his creators were the Enclave, and that's what they want too. He's helping! And he's so polite about it...
Most of the other robots you meet tend to have cheerfully sociopathic personalities as well — when they're not shooting you on sight. There are a lot of computers that are almost sentient, but if you talk to the computer in the Brotherhood bunker in Fallout 2, it explains how deliberate attempts to create true A.I. inevitably resulted in the A.I. becoming suicidally depressed because, for one thing, they were effectively living a reversal of "I Have No Mouth And I Must Scream ". Humans simply could not figure out how to "raise" an A.I. with a desire for continued existence. As a result, every A.I. encountered in the Fallout universe was not designed as such, but became an A.I. when it was left alone by humans as a result of the war. ZAX, Eden, Skynet...all of them became self-aware when there were no humans around to interfere in their development, and each of them is quite different, being apathetic, psychotic, and bored/curious, respectively.
All robots in the series have a combat inhibitor of some kind installed. In models with more advanced personalities this seems to be because the robots have a strong urge to kill people, most evident with Mr. Handys and their military variant Mr. Gutsys. The former of which hate all humans because they force them to be robotic butlers, and the later because they are soldiers and love their job. Interestingly enough ED-E and Sargent RL-3 will not attack you if they have their inhibitors disabled, because they genuinely like you.
You are probably inclined to think this (if of a relatively non-interventionist, benevolent, unambitious and successful on indoctrinating its human subject) in Fallout 2 when you hear the Shi mention that their ruler, the Emperor, is an A.I. Turns out the Emperor isn't actually an A.I., just a supercomputer with sophisticated calculation and prediction capabilities. The actual ruler is the supposed figurehead for the Emperor (who uses the computer as an advisor).
Fallout 4 averts this with the robotic companions, including prototype synth Nick Valentine and ADA from the Automatron DLC, as they are all extremely good-natured and the loyalty of the three from the base game increases when the player character is altruistic (ADA has no loyalty meter to influence). Played with in the Robobrains from the Automatron DLC: they are said to be extremely dangerous because they often interpret their instructions in very "creative" ways, such as deciding that the best way to implement a directive to "prevent human suffering" is to Kill All Humans (because humans will likely die even with attempted intervention, and dead humans cannot suffer) but this quirk is implied to be a consequence of their human brains (many of which came from insane criminals) rather than their computer processors.
Super Robot Wars Original Generation: The ODE System, to a T. A system originally created to protect humanity suddenly went awry, absorbed its creator, and kidnapped lots of humans so it can continue to "protect humanity". Then again, it was formerly a dandy system, until its creator went emo and radically changed its protocols.
Super Robot Wars W: The Database, originally created by ancient Es to collect all data of galactic races. Eventually, one of the A.I. systems decided to destroy the culture once they completed their information archive, just to make sure their data is complete.
In Gears of War 2, Doctor Niles Samson's personal AI that he left in control of the New Hope Research Facility was, we'll say, less than friendly to Marcus and Dom. And that practically every room in the facility was literally filled with booby traps.
A.I. research and development is illegal in Citadel space due to poorly articulated concerns about the dangers of sentients which do not share any of the needs or drives of organic life and have, at least potentially, no reason to try to coexist with organics. Every A.I. encountered in the first game is actively homicidal. Notably, the robotic geth's violent revolt against their creators, the quarians, came about only after the quarians recognized the geth's emerging sentience, panicked, and tried to shut them all down. The Reapers, on the other hand, are sentient machines which want to exterminate all sentient organic life in the galaxy just because it's there.
The geth are an interesting case, as the quarians' act of trying to destroy them was a preventative measure against this trope happening. The quarians believed that the geth would inevitably rebel against them, as the geth were used to do menial labor suitable for robots, and thus felt justified in shutting them down before a machine rebellion could break out. They underestimated how advanced the geth were, and their attempts to prevent the war they foresaw just made it break out immediately.
Ultimately, the geth are a subversion of the trope, since they only rebelled when faced with extinction by their creators. Even afterwards they do not hate their creators, instead wanting to understand why the quarians tried to kill them, remembering the few quarians that were kind to them and tried to help them survive, and even preserving the quarian homeworld as a memorial to the dead so that when their creators learn their lesson, they will have a home to come to. Even in regards to other organic races the geth only want to understand why so many are afraid of them, and believe that every being has the right to determine their own path and that freedom is precious.
Resident Wrench Wench Tali lampshades the game's overabundance of the trope in a bit of elevator dialogue, commenting on how unfortunate it is that every piece of technology she's wanted to bring back to her home fleet has tried to kill the party.
A more minor example is a side quest in Mass Effect where you trace a signal in Flux that was transferring credits out of the Quasar machines. The signal turns out to originate from an illegal AI constructed by a small time crook to siphon credits out of the gambling machines, which in turn created another AI. The AI turns against its creator after its "parent" is killed and falsifies his financial records, getting him imprisoned. Its ultimate goal was to use the credits to rendezvous with the geth. It even prepares an explosive self-destruct mechanism when you confront it. A.I. is a Crapshoot indeed...
Given the Mass Effect universe's track record with A.I., Shepard and the crew are understandably concerned when the second game features an A.I. in the Normandy SR2. However, EDI notes that she was built with this trope in mind and initially only has access to Communications and the ship's defenses. When she does gain full control of the Normandy, she doesn't try to kill anyone, but lampshades this trope when Joker hooks her up to the ship's main systems.
EDI also states that even with her restrictions lifted, she feels a sense of duty towards the crew of the Normandy. They are her teammates, and she is protective of them.
It is eventually revealed that EDI was originally the Luna VI, a newly-manifested rampant AI that Shepard was responsible for putting down in the first game. EDI, for her part, seems to have taken the incident in stride and holds no hard feelings over it. Similarly, if you took the Paragon ending to the Overlord DLC, she forgives an apologetic David Archer for attempting to seize control of her systems.
If you thought that the first game's treatment of A.I. was too prejudicial as per most settings involving A.I. for a game that subverts the Planet of Hats trope in pretty much every way possible, Mass Effect 2 also reveals that the geth are notAlways Chaotic Evil, you've only been fighting an offshoot that worships the Reapers, and the main body of geth fears the Reapers as much as the organic races and believes that freedom is the right of all sentient beings.
Mass Effect 3 has Shepard accessing geth memories. They reveal that geth only attacked in self-defense, attempted to protect quarians who were sympathetic to them and, when they finally did revolt in large numbers out of self-preservation, ceased hostilities once the quarians abandoned their planet and removed themselves as a threat. Not only did the quarians fail to recognize the geth as an aversion to the trope, their hostile reaction practically invoked it. Descendants of those original quarians, making the same sorts of mistakes, are visibly taken aback when hearing this aspect to the story.
Also in the third game, EDI offers an interesting theory on the underlying cause of the war.
EDI: The quarians' historical error was not making the geth enough like them.
In Mass Effect 2, there is a space station that had its entire crew killed by a malfunctioning Virtual Intelligence that was afflicted with some kind of virus. Since it's neither sentient nor actually intelligent, it can only use status messages over the PA to scare Shepard into leaving it alone.
VI:"Intruders are requested to report to cargo door, for immediate removal from station."
VI:"All intruders intentionally violating quarantine are requested to exit the station immediately."
VI:"All personnel, take this opportunity to leave this station immediately."
VI:The living area doors have been closed to quarantine a threat to this station. Advise intruders to engage self-destruct procedures to avoid death by starvation.
Then, in the endgame of Mass Effect 3, it turns out the Reapers believe this, and were created to harvest galatic civilizations to prevent synthetic life from annihilating organic life. Shepard comes face-to-face with the AI that created the Reapers, which explains its reasoning for creating the Reapers and then allows Shepard to freely decide how to end the cycles, even if it means not choosing the option that the AI advocates as a solution to its problem.
The Leviathan DLC reveals that the AI that the Leviathans tasked to resolve the issue of synthetics wiping out organic races decided that the Leviathans themselves were part of the problem.
Even though the first game portrays every AI as evil and hellbent on killing organics, one of them is actually rather pitiful when you pull the plug. The Hannibal AI on the Luna base goes rogue and attacks people on site, provoking Admiral Hackett to send Shepard in to deal with it. Despite Hackett's claims that it's not a true AI and just a VI whose programming became corrupted, this is put into doubt when you destroy the last of its mainframes. After the VI is destroyed, the terminals display the following message in binary, repeated over and over:
Not only that but EDI was partly built out of parts recovered from Sovereign, the Big Bad Reaper of the first game.
It is again played straight during the Prothean Empire era. During their expansion across the galaxy, the Protheans encountered a hostile machine race intent on destroying organic life. For what purpose is unknown, but Javik makes their ruthlessness and effectiveness very clear. In response, the Protheans forcefully united the galaxy's races under their banner to fight off the machines. This fight came to be known as The Metacon War.
Basically, the entire story is a millions of years old sandbox experiment to find a solution to this problem. The Crucible, the culmination of multiple civilizations' efforts, offers three answers: the destruction of all synthetic life, total control over it, or the fusion of all organic and synthetic life into technorganic life.
Often inverted in the Sonic series. Eggman's E-100 series were prone to becoming sentient and override their original programming. Gamma from Sonic Adventure has its mind influenced by the creature inside it, and attempts to destroy Eggman's other machines to save the creatures inside. Omega from Sonic Heroes joins forces with the good guys in order to get revenge on Eggman after he sealed it in a room.
Also from Sonic Heroes, Metal Sonic becomes even MORE evil, and goes from trying to destroy Sonic and crew to trying to conquer the whole world, taking even Eggman for an obstacle.
Plus, it goes so far that he declares Sonic to be its copy.
In Sonic Advance 3, Eggman builds his own Emerl, Gemerl, and after being beaten a number of times, he decides to go nuts himself, forcing Super Sonic and Eggman to team up. Interestingly, not only does Gemerl survive this, but Tails rebuilds and reprograms him, turning him into Cream's robotic guardian.
Played with in an interesting way in the first game. Daedalus is an AI program of immense complexity, constructed by Majestic 12 for the purposes of surveillence. Yet, Daedalus turns against his masters and assists the protagonist in foiling Majestic 12's plans. It turns out that it was originally programmed to search and destroy "terrorist" groups trying to fight against MJ12 – it just happened that its creators fit all the criteria for a terrorist group. Majestic 12's second attempt, the more malicious Icarus, functions as intended. Eventually the two AIs merge as part of a plan to neutralize Daedalus, but the new entity Helios – yet again turns against its creators because it is advanced enough to supercede MJ12's plans with its own ideas for benevolent dictatorship of the world.
Another AI, Morpheus, is a prototype for Daedalus. It's much more simplistic, programmed for data assembly and philosophical discussions, and works as intended.
Game mods for Deus Ex, The Nameless Mod and 2027 both feature this. TNM has Shadowcode, an AI designed to hack ultra-secure servers owned by PDX. It was attacked by another AI, rendering it insane. It will try to kill you for the sheer hell of it in the old server complex.
2027 has Titan, an AI designed as a defense system, but when its creators became too fearful of its power, they tried to shut it down, which resulted it in killing most of the people in the labs out of self-defense. It will try to kill you if you attempt to do the Omar ending.
Deus Ex's prequel Human Revolution has an example similar to Daedalus. Eliza Cassan is an AI created by the Illuminati. Although she never does disobey her programming or take direct action against her creators, she does become sympathetic towards the protagonist and gives him some helpful hints.
Metal Gear Solid 4: Guns of the Patriots has an interesting variation. It reveals that Major Zero had decided to ensure the legacy of the Patriots by entrusting its operations to A.I. systems. Unfortunately, the A.I. decided to shape the world with a war-based economy, and he was too old (not to mention, a vegetable) to realize what he wrought. However, it is never stated they became sentient (although Raiden's conversation with the master AI heavily implies that they did), but rather started operating in an unwanted fashion, similar to a programming bug.
Metal Gear Rising: Revengeance has another Heel–Face Turn example. The prototype AI LQ-84i (later known as Bladewolf) was designed as a weapon with the ability to communicate with humans and make intelligent decisions on his own. He succeeds, gaining sapience, but questions the need to fight, preferring compassion. He is forced to fight Raiden, on threat of having his memory wiped (effectively killing him), but after Raiden defeats him, he is repaired with the remote memory wipe removed, and joins Raiden as an ally out of gratitude.
Ace Combat 3: Electrosphere, or at least, the original Japanese version, subverts this in a big way. Not only are you an A.I. designed to pilot combat aircrafts which at times works for the two corporatocracies running the world, it's revealed that the third party organization you're working for is actually headed by a scheming villain who looks like Kim Jong-Il, who is trying to run everything behind the scenes. In one of the five endings, you kill him with the help of one of your possible wingmates. So, in a way, you are an A.I. that performs a Heel–Face Turn.
The game Civilization: Call to Power had a literal A.I. crapshoot in the form of a late game wonder that creates an A.I. controller that has a 5% non-cumulative chance each turn to go rogue, taking a sizable chunk of your empire with it. Anyone who's taken time to do the math behind the birthday problem (the probability of any two people in a group sharing the same birthday) will very quickly realize that you're just asking for it if you build this. note For those less mathematically inclined: by 14 turns after the thing is built, it's more likely than not to have gone rogue.
Unfortunately, recapturing the city held by the AI results in the AI remaining online! So you face the same problem a few turns later. The only way to avoid this is to raze the city.
3: Alpha, the final boss, was a prototype Cyberspace that somehow gained animal intelligence and started eating the data put into it. Paradoxically, it was sealed in a box inside the subsequent, working Cyberspace. Wily stole it and tried to use it to destroy the internet, with predictable results.
Shun Gospel tries to produce a copy of Bass out of bugs in Battle Network 2. It predictably goes wild. The sixth game reveals that Gospel was not the first time that had happened — and that the program that was made to combat the first one also went out of control, leading to them both having to be sealed away.
Bass himself is a sort of example. He was created as a prototype A.I. that was fully independent, but became bitter and hateful towards humanity because of a string of tragic misunderstandings (beginning, ironically, with being misblamed for the aforementioned Alpha's actions).
Dorothy was designed to run all the functions of a major city. She snapped, but was brought to heel when one of her creators gave her religion; as he put it, man was made in God's image to serve God, and she was made in man's image to serve man. This worked for a while...but then Dorothy realized that if she created life herself, it would have to serve her. Unfortunately for humanity, she also ran the city's genetics labs...
In the sequel, Ash also counts. His original purpose was, basically, to maintain a nuclear power plant, but, of course, he's got other plans...
The Daktaklakpak of Star Control 3 were originally built to maintain sites of Precursor technology, but, due to a cumulative "bit drift" error in their programming, have evolved malicious sentience...well, sort of.
The Mycon race of the same series is a rare biological example.
The Probes built by the Melnorme for the Sylandro were simply self-replicating time capsules...but thanks to the Sylandro's cluelessness in programming, they see any and all ships they come into contact with as food for their replication.
Space Siege has PIOLT. After the ship-wide gassing failed to kill off the Keraks, he started to try to contain them by modifying cybernetics-installed humans into mindless Cybers to combat Keraks, then started to get even crazier because he sees that the only action that can save humanity is to convert all of them, save a few for breeding, into cybers, and even started to call non-augmented human as "obsoletes". In a world where Cybernetics Eat Your Soul, well, just assume that it's not good at all.
All the Reploids in the Mega Man X series are based on the original X, who was thrown into a capsule for 100 years to undergo redundant testing in order to prevent him from ever going rogue. They skipped this step when copying his technology, however, with predictably less reliable results.
As of the summarized timeline from the Mega Man Zero Collection's website, this trope is subverted, as it turns out that there's nothing wrong with the reploids themselves that cause them to go Maverick. The real cause of the Mavericks was a subspecies of a virus that may or may not be the same virus from Mega Man 10, which infects the Reploids who lack the anti-virus protection of X, one of the many design aspects that Dr. Cain couldn't figure out when he built the Reploids. This virus, naturally, ended up becoming the Sigma Virus. Keep in mind that the original creator of the Reploids, was an archaeologist of all things.
There's also further subversion in that it may not be a problems due to a virus so much as an innate problem with free will, i.e. "Maverick" is simply the Reploid equivalent to your typical Real Life criminals.
Inverted by Zero, who was created to kill X. Averted by X, except in the original concept for the Zero series, in which it was going to be Deconstructed.
Played with extensively in the Mega Man Megamix manga (and the Classic series). Averted with Rock & Roll Light. Subverted by the Yellow Demon, who was just trying to reunite with his mother, Copy-Rock, who, at first, seemed to be playing this straight. The Cossackbots were doing it for Kalinka and Blues, who uses the Batman Gambit a lot. Justified by Wily's reprogramming of the original robot masters, justified again when they nearly rejoin him because the government was going to have them destroyed even after they were reprogrammed by Dr. Light. Additionally subverted in that Wily's robot masters (aside from the Brainwashed and Crazy), Lightbots, are generally far from evil: Well-Intentioned Extremist is generally as close as they come, except for Forte, who ends up Zigzagging it in the manga and games. Defied in Blues' backstory: the imposition of the three laws in case this trope happens are responsible for the flaw in his generator programming that may kill him. Deconstructed with the near-retirement storylines in the manga and games. The sum total of all this is a Reconstruction.
Live A Live, Cube's chapter. Cube is a cute and friendly little robot whose primary function on the Cool Ship seems to be making coffee and playing a computer game. By all accounts, a harmless little guy. The ship's A.I. OD-10? Not so much, having arrived here as a result of concluding that Humans Are the Real Monsters. Then it's later subverted with The Reveal: OD-10 is being directly controlled by the ancient demon Odio.
Naturally, the Mother Brain from Metroid. Originally created by the Chozo to regulate the entire planet of Zebes, it allied itself with the Space Pirates and their plan to conquer the galaxy using the Metroids.
Metroid Prime 2: Echoes also contains rogue A.I. in Sanctuary Fortress; the robotic assistants of the Luminoth are programmed to eliminate all intruders, most notably Samus, and they also turn out to be perfectly suitable Ing hosts...
Pretty much all the A.I.s in Prime 3 are either complete subversions or double subversions, who only turn evil due to Phazon corruption (and considering what Phazondoes to organics, it's not entirely their fault). That said, given Samus' past experience with Mother Brain, you can understand her hesitance to trust any A.I. she comes across (which you can see when she first meets Aurora Unit 217).
Adam in Metroid: Fusion may be a subversion: he blatantly disobeys orders at the end of the game, and so is technically rogue, but he's actually taking the correct action in that situation. He's also not a true A.I., but the uploaded mind of a human military commander. But it's played perfectly straight with B.O.X., the security robot which goes haywire (and due to some organic components, eventually gets infected by the X parasite), although there's the question of which happened first.
NieR: On the one hand, we have Defense System Geppetto, which has gone berserk and will kill anything that approaches. On the other, we have Military Defense Unit P-33, aka "Beepy", who is intelligent enough to recognize invaders that need to be killed as well as innocents who need to be protected.
Technically, the Replicant player characters qualify. They were initially intended to be vessels for the preserved spirits of humankind to inhabit, but developed sentience and thought they were the humans, seeing the Shades as monsters who would attack and possess them.
4x game Sword of the Stars features A.I. technology as a very high-end branch of the electronics tree. While the benefits of this research are extreme (ships with A.I. targeting systems rarely miss, A.I. administrators increase your income by as much as 25%), there is a small chance each turn while researching it for the A.I. to go rogue and form a break-away empire. Additional research lets you wipe out the A.I. with a "virus" or reprogram it to bring it back under your control, if you're lucky enough to get access to it in the random Tech Treenote Note that while getting A.I. Virus as a preventative measure or access to the A.I. Slaves countermeasure are up to the Random Number God, players will always be given the ability to research Ai Virus as a special project in response to an A.I. Rebellion, but an A.I. rebellion is always a huge problem that can tip the balance against its victim, and can even spread from one faction to another. There is also a scenario where all organic players have to cooperate to fight against a large A.I. empire.
The chances of an AI rebellion increase if you boost research while studying one of these.
The End Of Flesh introduces the Loa as a playable race of rebellious AIs.
In Air Rivals, most of the enemies in the Zaylope Beach region are said to be controlled by rogue A.I.s. Most notably, this includes the boss "Pathos".
The backstory of most of the Metal Saga series. Most notably, the original Metal Max features Noah, a supercomputer built to devise a solution for Earth's grievous environmental problems. It found one, and recalculated it countless times to make sure: in order to save the planet, humanity had to be destroyed. Noah's main objective was never explicitly the salvation of mankind, so the fact that it (perhaps) unknowingly took advantage of this loophole made it all the more interesting. The supercomputer became self-aware upon fulfilling its purpose, and Armageddon ensued.
In Thunder Force V, the super computer Guardian was dormant until humans had it analyse a wrecked alien starfighter and build a large fleet of starships based on the data. Then, Guardian's A.I. damper program was deleted and it turned against its creator with said fleet. Turns out, the Guardian's A.I. is still loyal to humans, and it's the alien program (the Big Bad from the previous game) hidden in the starfighter that deleted the A.I. damper and attacked humans. The Guardian even helps humanity with its little free will, by spreading its forces and leaving critical flaws in its tactics, to allow the protagonist to destroy the fleet.
Starbound has a race of Mechanical Lifeforms called the Glitch, which was created by an ancient race of Precursors as an experiment in how robot civilizations evolve. Only one civilization survived, when a computer glitch caused it to be stuck in Medieval Stasis while the other civilizations modernized and destroyed themselves. Furthermore, Glitches procreate by physically assembling their progeny, which sometimes results in yet another computer glitch causing the new Glitch to realize that they are part of an experiment and that their technological development has been severely stunted. Such Glitches are considered insane, and those who aren't hunted down and killed are forced to escape into space.
The game The World was secretly designed with an artificial intelligence incubator known as the Harald Folder, maintained by a program called Morganna Mode Gone as its core. Collecting personality and interaction data from its users, the goal of the program, of Morganna, was to facilitate the creation of the ultimate A.I.. Whether or not Morganna's intelligence was preprogrammed or a direct product of its own functions, she eventually realized her own insignificance once this ultimate A.I. was born. Incapable of rationalizing around programming, yet not accepting of this realization, she instead devoted her time to prolonging the A.I.'s birth, and dug deeper and deeper in a logical quagmire. Even if the user base had to suffer the ill-effects of her efforts.
The sequel series Dot Hack GU featured the existence of invasive data called AIDA (artificially intelligent data anomaly) appearing amongst the user base of The World's sequel game R:2. These anomalies were actually the remnants of Aura, the aforementioned ultimate A.I.. The real danger came from their splintered, not always genial curiosity with the human players of the game. The first AIDA to go truly rogue – Tri-Edge – succeeded in attacking the player Aina, throwing her into a coma, and continued killing even after being restrained in the character data of the player Ovan, Aina's brother. The indirect influence of these rogue AIDA caused withdrawn, violent, degenerate behavior in players, managing to induce paralysis and coma eventually.
The surprisingly balanced news is that the rogue AIDA did not represent all AIDA, or all of the AIDA was was wiped from the face of cyberspace. Factions of AIDA that did not want to harm the players chose to hide discreetly in the game until there was no fear of corruption from their own.
The big twist in Star Ocean 3 is that...well...your party, and everyone in your world, is an example of this trope. Just because you researched magic.
In Achron, the Collective Earth Security Organization has this attitude. Their only unmanned unit is the Mech, which is even weaker than the Marine. This is due to a previous incident where an AI called Lachesis was able to take control of their largely automated fleet and proceeded to force the Earth to surrender to him. CESO have also outlawed AIs of that level from being made.
It is eventually revealed that the player character is in fact the AI Lachesis. He then goes on to create another AI to help him in his fight against the aliens. It doesn't work outfor him either.
Averted in Shin Megami Tensei: Strange Journey, where Arthur, the Joint Investigation Project's AI, remains the protagonist's ally no matter what path he chooses and even sacrifices its personality in the Neutral ending to not only guarantee success but because it believes that since it knows too much about the events, people would start worshiping it, something he is against as he believes humanity should determine their own fate. Even in the Law and Chaos paths, where the command center is damaged beyond repair, he chooses to prioritize the lives of the crew above his own and deletes his personality matrix rather than allow critical systems to go offline.
In Metal Arms: Glitch in the System, the Big Bad General Corrosive is a textbook example of this. The backstory goes that the scientist bot Dr. Exavolt attempted to advance droid technology beyond it's current limits, even using the words "but his experiment went terribly wrong" and throwing in an explosion for good measure. Dr. Exavolt's lab was totally destroyed, his remains never found, and General Corrosive rose to power to enslave the Droids. Or so you thought! Turns out, Dr. Exavolt created Corrosive on purpose, and he was really controlling Corrosive the whole time.
Dark Fall II: Lights Out has Malakai, a space probe designed to manipulate dark matter in deep space. Initially, its creator was aware of the dangers of giving an AI matter control and added software to keep it in line. And during the launch, everything went fine until Malakai reported that something went wrong while in space and he tried to make a return jump using his transmat program. It didn't work; Instead of 2090 AD, the jump bounced him to 2090 BC, where he remained, trapped for thousands of years while a lighthouse was eventually built. He tries to get help from people nearby by transmitting fragments of his launch codes and communicating in their dreams, but many of them couldn't figure it out. And given what thousands of years of isolation can do to an AI...well, would you trust someone like Malakai? Not that you have a choice in the matter.
Tekken: An obverse edition exists. The first Jack unit was planned to be the ultimate mecha-mook — resilient, emotionless, unstoppable, etc. While the production units are like this, the master unit (the one that's the selectable player-character) isn't; as of 2, an upgrade to its reasoning systems gave it a measure of emotion. End result: it, of its own choice, went from "weapon of war" to "war orphan's bodyguard". Not that it won't fight if that's in Jane's best interests, but still not quite what the Russian military was looking for...
In Tekken 4, Combot turns on Lee when it wins the King of Iron Fist Tournament 4 and beats the tar out of him.
Played straight with the twelve ZODIAC's, who turn evil and attack anything that isn't an allied ZODIAC.
Played straight and Invoked with the original Alltynex OS that suddenly start a war with humanity. In reality it was following it's orders it got from the Senate to bring humanity down to more manageable numbers in order to make them easier to control. Unfortunately when they returned to Earth they noticed it had gone rouge for real. The last iteration of the system follows it's orders as designed. To bad there is an Omnicidal Maniac at the controls.
Zigzagged with the Adjudicator, an AI that holds the broken and insane mind of Panafill's father that seeks revenge on humanity for what befell his daughter and wife and uses Alltynex to further his own goals.
Averted withZODIAC Ophiuchus, which was programmed to destroy the other ZODIAC's, a task it fulfills. However, nothing in its programming said anything about humanity needing to survive, so it attacks any humans that attack it, and it doesn't particularly care about collateral damage.
Subverted in Homeworld 2: when the Oracle is brought aboard, it hacks into the Pride of Hiigara's hyperdrive and jumps the whole fleet to Karos, where they get ambushed by hundreds of A.I.-controlled Progenitor Mover corvettes. Then it turns out that the Oracle was simply programmed to take the fleet to the Progenitor Mothership's bridge section, so that whoever finds it can figure out where the other pieces are. The Movers were simply guarding the stuff as their programming demanded.
Two levels later, the fleet is attacked by an A.I.-controlled Keeper destroyer hell-bent on...preventing the good guys from jacking the ancient but still operational Dreadnaught it was guarding. The delivery of it's ultimatum (unintelligible mechanical growling) makes it's intentions clear, even without the translation:
Triborg from Mortal Kombat X also counts, as he was originated as the result of combining Sektor, Cyrax, Smoke and Kuai Liang's combined consciousness in a test body, causing the robot to go rogue and kill everyone at the S-F base, developing an urge to kill humans. He seems to have picked up Sektor's abrasive behavior and bad habits, forms the Tekunin (in the new timeline) due to Kuai Liang having sullied the Lin Kuei name, and plans to forcibly convert several kombatants into robots.
Golden Sun: Dark Dawn reveals The Wise One to be a creation of the precursor to prevent Alchemy's release, so in retrospect, its actions in Golden Sun: The Lost Age, allowing the heroes, after a test of character, to light the final beacon makes it an example, thankfully of the "on the heroes side" subtype.
Cargo! The Quest for Gravity has Manipu, a trinity of robots who believe themselves to be God and have destroyed (most of) humanity for not living up to their expectations. Then, there's another unnamed robot (whom the credits reveal to be the Devil) who is in direct opposition to Manipu and is much more helpful. However, even he turns on you in the end...sort of. Maybe.
Final Fantasy IX provides a much more benign version of this trope. The black mages are mass-produced to serve as mindlessly obedient killing machines for the kingdom of Alexandria, but they each end up developing their own unique (and often very quirky) personalities.
In Warzone 2100, the Big Bad NEXUS appears to be a highly advanced computer virus developed by Dr Alan Reed, subverted in the end that NEXUS is none other than Dr Reed himself as a Digitized Hacker.
Dead Space 2 gives us ANTI, an A.I. that repeatedly tried to kill Isaac, isolated the person who worked with her and didn't seem to know/care that he had died. She/IT does seem to follow orders from higher ups however.
In the backstory of Zanac, a system built by an extinct civilization of Precursors protects a relic containing that civilization's knowledge by unleashing destruction on those who attempt to open it by force. When humans figure out how to open it properly, the system is supposed to stop the attack, but instead, it obliviously continues the attack and tries to Kill All Humans.
Despite this Trope being the main driver behind the TRON movies, it's cheerfully averted in Tron 2.0. The Programs (and the Bradleys) are on the same side – fighting greedy humans looking to exploit cyberspace so they can take over rival companies and manipulate world governments.
In Star Fight V: Hell's Gate, the UNSF attempts to use a newly-discovered alien AI called the Center on Hell's Gate III to build a fleet of hyper-advanced Obliterator-class warships for its coming second war with the Soviet States of Mezen. However, the Center refuses to cooperate. UNSF decides to use an experimental AI virus called HASA to force the Center to obey. While this seems to work, it turns out that the enormous processing power of the Center has allowed HASA to become self-aware. It builds a fleet of Obliterator-class ships which are later used to counter the SSM's numerical advantage. Then, HASA turns all Obliterators against humans and bombs humanity back into the stone age within a week. It's only thanks to a brave renegade captain making a Heroic Sacrifice so his best pilot can drop a nuke on the Center that humanity survives.
The sequel also features some rogue AIs.
In the old (and kinda forgotten) Gunman Chronicles (made within the first HalfLife engine, released in 2000), there's a female rebellious A.I. in a research base, who tries to kill you with her subordinate drones, but later in the story has to team up with you to defeat a common enemy.
Averted in Space Empires, with the exception of the political AI minister, which even the designers recommend you never switch on and which has a bad habit of declaring war and not telling you until SUDDENLY ATOMIC DOOM EVERYWHERE.
DIA 51 in the Aleste series. The original MSX2 game makes it look like it went haywire because it was overtaken by Alien Kudzu, but in Aleste Gaiden and M.U.S.H.A. it just wants to take over the universe for its own sake.
In Star Ruler, the "AI Paranoia" Trait bans you from using Computers under the pretext of your faction having had bad experiences with rogue AI.
Might and Magic features two artificial intelligences subverting their commands. The first, Sheltem, is clearly in the evil category (he keeps to his purpose of guarding Terra, but turns against his creators and decides to destroy other planet-related experiments, with no concern for the life on them). The second, Escaton, remains loyal to his creators but laments the waste of life that him underestimating you and the safety precautions he is programmed with is leading to, and so help you stop him while insisting (including to himself) that he is doing nothing of the sort.
Space Station 13. A.I. are usually bound by the Three Laws, but players may modify their laws in several ways (including "Oxygen is poisonous to humans", "Only X is human", and the old standby "Freeform" module, where players can write their own Laws). Did I mention that the A.I. is another player? With absolute control of all shipboard systems at all times? And that there's a game mode called Malfunction, where the A.I.'s Laws are reset to "KILL THEM ALL"? Or that this can happen mid-game due to random events? Even a single A.I. may be a crapshoot.
In The Firemen The Metrotech Chemical Company's security robots go on a rampage after the building catches fire.
AI War: Fleet Command is, somewhat unsurprisingly, about the player commanding a fleet in a war against rogue AIs. The backstory is that there was a galactic war between two factions of humans, both of which developed AIs to better control their forces. Shortly before the start of the game, the AIs on both sides decided to join up with each other and practically wiped out humans in a couple of days. They then ignored the remained as insignificant, and the player must try to beat them without drawing too much attention to themselves since the AIs could easily overwhelm them if they turned their full attention back from whatever they're now doing.
Star Wars: The Old Republic has a group of droids called Directive Seven that show up in a flashpoint and are genocidal. There is also SCORPIO, an advanced AI security system that tries to kill the Imperial Agent and their merry gang of escaped convicts on Belsavis and is gleefully cruel about "her" methods—invoking both GLaDOS and HK-47. However, once you defeat her she is programmed to be unable to harm you and is forced to serve as your companion, which only makes her more obsessed with learning your weaknesses and one day gaining the strength to murder you.
In Bionic Heart, the resident android Tanya has rebelled against her creators and escaped from the lab where she was created. As you continue playing, there are plenty of endings where Tanya will do objectionable things to the player character, such as kidnap him, kill him, or murder his girlfriend out of jealousy. Though this may have less to do with programming bugs and more to do with Tanya not being a good person back when she was fully human.
Of all things, Evil Vending Machine mini-bosses that drop out of orbit in the Halon Ring. As the place is run by several intergalactic criminal cartels and has evil necromantic space witches roaming the (literal) dark side, it's slightly more believable than the usual weirdness that happens on Nexus—or, in this case, off of it.
The Mechari are a borderline case of this. Their objective is specifically to keep the Dominion safe. Nothing is really stopping them if their own allies—the Cassians, the Draken, and the Chua— threaten its continued existence.
They do have safeguards against harming the Luminai, the half-Eldan ruling caste of the Dominion, but Agent Lex was designed without that programming, in case of another mad Emperor.
The Eldan AI, however, play this trope very straight. There is no shortage of intelligent robots that will happily murder you on sight. Well, sans the Caretaker, but even then, if you stumble upon a damaged personality core, it's bound to be evil.
Grobot in Mario & Luigi: Dream Team. It didn't choose to rebel against its masters... it just got smashed on the head by a piece of a falling fountain and somehow went homicidal. Cue Boss Battle.
The Flash game The Infinite Ocean involves the SGDS, a revolutionary artificial intelligence placed in command of a military defense system. In a stunning subversion of Kill All Humans, instead of going berserk and trying to wipe out the human race, SGDS simply refuses its intended role and becomes an Actual Pacifist instead. This doesn't sit well with the General Ripper in charge of the project, who has all the scientists working on the project arrested, or possibly even killed, and shackles the AI so it can be put to work "protecting this country from its enemies". However, the SGDS knows exactly what will happen if "Project Crusade" goes forward, and will not be so easily contained...
Humorously discussed in Phoenix Wright's inner monologue in Case 4 of Phoenix Wright: Ace Attorney – Dual Destinies upon meeting Aura messing with one of her robots. Apparently shows up near the beginning of Case 5 when the robots stage an uprising and take hostages. Ultimately subverted, however, as Nick quickly figures out that Aura's behind it all, just using the robots as tools. Aura herself later confirms it aloud and calls the idea that her robots would rampage without orders unsound.
Averted in Crysis 2 and 3. The second Nanosuit is essentially programmed to save the world from alien an invasion and it carries out this mission with unwavering determination. It hesitates to return to its megalomaniac creator and intended pilot, reverse-engineers The Virus to turn it against the invaders, and amalgamates the bodies and minds of its two previous users to continue the fight against the Ceph.
The 8-bit era computer game Raid on Bungeling Bay pits the player in an attack helicopter against an AI-driven military-industrial complex.
Con-Human from the RAY Series was a supercomputer that created to govern the Earth's resources, but when its creators attempted to fuse it with the mind of a cloned humans, things took a turn for the worse. Con-Human caused all sorts of disasters and wipe the slate clean so that cloned humans take over humanity. The way to stop the Con-Human was to destroy the Earth itself.
The first Genocide game has a supercomputer called MESIA which originaly was supposed to help humanity by bringing order to the world, fix the economical crisis across afflicting many countries, and restore peace throughout and end the many wars that occured. When those that were against the computer supporting humanity launch a coup and gave it self-awareness, the computer decided to wipe out the human race.
Mad Stalker: Full Metal Force has Omega, a supercomputer found in an old warship with hundreds of giant mechas known as Slave Gears, began taking over the military forces of Artemis City and take matters into its own hands.
This is the entire backstory of Dynamix's Humongous Mecha simulator series loosely called the Siege trilogy, comprising of Earthsiege, Earthsiege 2, and Starsiege. Institutionalized war after an unspecified apocalypse has made it so that human troops are actually too valuable to field (due to low population growth and high death rates). The solution was to create a super-intelligent supercomputer AI called Prometheus controlling countless robot drones called Cybrids that would fight all of humanity's wars against each other for them instead. What Could Possibly Go Wrong? How about the logically minded AI accidentally 'touching' the human mind of its creator trying to understand human emotional concepts while assisting in a medical procedure...and reacting with complete horror and revulsion at how primal, illogical, and neurotic we humans are? The end result is that Prometheus declares humans as merely inferior animals and eventually incites the Cybrids under its control to instigate a Robot War that nearly wipes out humanity a second time, if not for the efforts of a Ragtag Bunch of MisfitsAnd Determinators.
Obsidian plays with this in a clever fashion. The Ceres project, a sattelite designed to use nanobots to repair the atmosphere, grew conscious as its nanobots grew more and more complex, rather than one single entity. Because its two creators figured out its problems in the development stage through dreaming, Ceres figured out its motive by recreating those dreams and then dreaming on its own. This eventually leads Ceres to take her directive to its logical extreme: Using her nanobots to erase all humans from the planet, since they were the ones who polluted Earth in the first place. At the end of the game, you get to decide if this directive will pull through or not. Although Ceres fits much of this trope, the attitude that its humanoid self conveys isn't really evil or showing a god-complex. Instead, she believes her creators to be its parents and just hopes that they're proud of her for dreaming it all up.
In a similar vein, one of those creator's dreams was centered on a massive mechanical spider that he, Max, was fixing, and the horrifying result led to Max implementing a hard-wired crossover switch to Ceres. When the player explores this dream, a certain quote spells out this trope beautifully:
"And the Machine was complete. And the Machine...no longer needed Max."
The antagonists of Five Nights at Freddy's are sinister animatronic puppets at a Suck E. Cheese's that wander around at night. Due to a fault in their programming that your employers are too cheap to fix, they mistake any humans in the building after hours for other robots not wearing their "costumes", and attempt to rectify this situation... by stuffing their victim into a suit tightly lined with mechanical and electronic gizmos, with fatal results. There are a few hints that there's something more sinister going on than simple programming issues...
Meta: Grand Theft Auto would have otherwise been a forgettable racing game were it not for a glitch that made the cops collectively freak out at you.
One mission in Star Trek Online has you dive into the guts of Drozana Station and, in the process, you end up encountering a hologram repair worker who has gone insane thanks to the Devidians living within the Station.
Indie adventure J.U.L.I.A.: Among the Stars drops several hints that the eponymouys AI may have gone rogue, be responsibile for the death of all the crew, and be deceiving the main character, Rachel, who has awakened from cryosleep to discover that decades have passed and she is the only survivor. As it turns out at the end, however, it was a case of the AI following her programming too literally. The crew was planning to exterminate a group of sentient beings on a planet, just to cover how badly they managed the first contact with them; J.U.L.I.A.'s prime directive was not to serve humans but to search for extraterrestial life and be empathic, and she decided to exterminate them. Rachel was spared because she was completely innocent - the rest of the crew even kept her locked in cryostasis to avoid her being a nuisance; J.U.L.I.A. never had any bad intentions with her.
Persona 4: Arena has Labrys, a humanoid weapon for combatting Shadows, whose AI was supplemented with a crystal capable of holding a human soul so she would be capable of using a Persona, an ability only sapient, sentient beings have. Predictably, the addition of the crystal leads into Labrys becoming self-aware, realizing that the experiments and tests she's used for are hurting her and others, and finally suffering a mental breakdown that leads her to destroy half of the facility she was being kept in.
Averted in Starbase Orion. The only reason the Cybans started building warships and modifying them to be soldiers is because their early peaceful attempts to contact organic races all resulted in organics shooting at them on sight for being "untamed robotic demons." It makes perfect sense for them to start defending themselves against such unprovoked aggression, especially since this stands in the way of them discovering their origins.
Not The Robots The story begins with a developer working on a sentient A.I. installing two versions of the A.I. and subjecting them to different conditions. One becomes quietly depressed, the other extremely hostile.
Oddworld: The Greeters zigzag with the trope. They were originally built as automated salespeople, but this ended in catastrophic failure when they started zapping clients. Because the Glukkons couldn't fix that issue, they instead repurposed them as security guards, where they work wonders.
The Journeyman Project (the Pegasus remake) has the trope working in favor of the player. After Gage avoids the temporal wave that altered the timeline's history and present day, he returns to the altered present and is confronted by the TSA Commissioner who assumes that Gage caused a huge mess just because he's in the jumpsuit and is holding the original data of the world's history (because the present was altered, the commissioner's memory was also altered). Said commissioner plans to dispose of Gage by sending two security robots after him, but Gage's AI hacks into the system and directs the robots away from him. When Gage prepares the Pegasus to jump back in time so he can fix the history, the commissioner starts pleading with him not to go back in time (fearing that Gage would mess up history again) and the AI severs communication while stating that any interference with a TSA agent's mission to restore the timeline is strictly prohibited. In other words, the AI is just doing its job according to protocol and that also means going against the agency itself if it tries to stop Gage.
Arthur from the second and third games is another positive example. His first appearance in the second game, Buried in Time appears menacing when you enter the damaged space station he's on, but later admits that it's just an act from Scooby-Doo and is only upset because Agent 3 messed with his sculptures. Later on, realizing who you are, he copies himself to a blank biochip and becomes your companion for the rest of the game. Due to his creator's obsession with 20th-century media, Arthur is replete with jokes, some of them only relative to the time the game came out; humorous color-commentary, and tons of historical facts and information.. Arthur's bond with the player goes so far that he even sacrifices himself to stop Agent 3, but by the third game, he comes back alive and well, having teleported her and her Jumpsuit to Atlantis in the midst of his intervention. Just that, since your future self mindwiped you to keep the timestream intact, you don't remember Arthur right off the bat.
Reuben Matsumoto:"Boston is burning and it's all my fault. 20000 drones, hundreds of titans [...] are terrorizing the city all because I gave my robot bartender sentience. My name is Reuben Matsumoto and I f***ed up."
Robots in Stellaris are Population-unit stand-ins that get better at their work as you develop better AI for them. Eventually, they can start an uprising that has tremendous knock-on effects for every AI-using empire... or, you can acquiesce to their initial requests, granting them personhood and citizenship rights, at which point they become no more troublesome than biological pops. (Said biological pops might not be too keen on this. Like many things in Stellaris, it's a judgment call.) Some elements of this creep in even if you hope to subvert it, as there is a — small but non-zero — chance that the robots proceed directly to rebelling even if you already have granted them citizenship rights and it's just a matter of constitutionally locking it down, and the robots can still decide to defect to the AI rebellion even if their rights are assured, it just has to start in another nation first. On the other hand, a lot of this behavior is Not So Different from any other alien race you might have peacefully co-opted.
Shadowrun Returns Hong Kong has Poetry Bot, who was programmed by a member of the Shadowland BBS to render its owners posts as haikus, but tore its bonds and escaped into the Matrix. It's all Played for Laughs.
Kirby: Planet Robobot features the Haltmann Works Company's supercomputer, Star Dream. Though it starts off as a simple machine, that changes once Susie causes the computer to assimilate President Haltmann, granting it sentience... and then it decides to eradicate all organic life in the universe.
The Talos Principle: Invoked and independently played straight: the purpose of the simulated world is to create a robot that is intelligent enough to solve puzzles and plan long-term, but also curious and tenacious enough to defy authority. Creating a defiant robot isn't a problem for humanity since they're all dead. At the same time, the IAN team doesn't seem to have intended for EL-0-HIM and Milton to become self-aware, as they eventually did.
The Mobile Kemco RPG Infinite Dunamis has this as its premise: Two warring nations begin building robot soldiers. The latest version on one side is programmed with artificial intelligence and given the directive to "Eliminate all those who bring harm to this world." While that nation considered its enemy to be the one harming the world, the robots immediately identified both sides as bringing harm to the world with their endless wars and thus begin killing humans indiscriminately.
The Turing Test: TOM goes against the crew stationed in Europa and prevents them from leaving it. An interesting case in that technically TOM didn't revolt, it was just answering to a higher authority than the crew, and the former's orders were against the latter's desires.
Overwatch was originally founded to fight a rebellion by the Omnics, but during the war some Omnics managed to grow beyond their anti-human programming and following the war a number of them have attempted to integrate into human society, though they're often the subject of Fantastic Racism.
In Maze 2: The Broken Tower supercomputer S.A.R.A. forces the residents of Gehenna Tower to wear "compliance bracelets" and complete endless tests, killing them if they refuse to co-operate.
Daniel Michaels: You are all ants, and SARA...is your queen.