AIs shutting down sec because of law 1

[Luke Cox]

I get that Asimov AIs have to try to prevent executions and harm to prisoners, but for the love of god can we please make it a rule 1 violation to bolt in every single security officer and cripple sec if the AI decides that somebody in sec caused harm?

[lzimann]

You are talking about the round that just happened, right? It was not a simple harm, security killed like 2 people, if that's not enough for an ASIMOV AI bolt everything, I don't know what is.

[Cik]

the response should always be proportional

the response to 1 securitymen harming 1 guy is very different from the response required from multiple security guys harming multiple people and clear radio chatter confirming that they are unrepentant and will continue to do it

i escalate linearly along a certain path

at first sight of harm i talk to the individual officer via holopad, if he is available or the HOS. i secure a promise that they will not do it again.

if harm continues, bolt the harmer.

if harm spreads to multiple, bolt multiple. ask again for promises not to continue.

if no promises are given or security is unrepentant and/or vows to continue send in the secborgs and take over the brig as best you can.



FYI security law 1 requires the AI to bow under to hostage situations. take a prisoner and threaten to kill him if you really need to get out.

don't expect any favorable treatment after that, though.

the key when playing AI is to respond PROPORTIONALLY and leave your adversary a way out by being a nice human and cooperating. NEVER put their back to the wall unless you absolutely must, as it will usually either get them or you killed and lets the station descend into anarchy meanwhile.


i don't know about policy but maybe i'll write a "ai tips" guide or something.

[Luke Cox]

You are talking about the round that just happened, right? It was not a simple harm, security killed like 2 people, if that's not enough for an ASIMOV AI bolt everything, I don't know what is.

In that round, I executed one guy in the prisoner transfer room, then a borg walked in for no discernible reason and immediately stunned and cuffed me after the guy was already dead.

This is something that happens all the time though. If anybody dies in sec for any reason, all of security gets shut down while antags murder freely outside. It's a huge pain in the ass for everybody and AIs get away with it constantly. An appropriate response for the silicons would be to remove the harmful items involved and take the wounded to medbay, not cripple an entire department.

[Zilenan91]

just remove secborgs tbh fam

[Steelpoint]

Reasonable action should be made only on the offender, and not the collective. Too often I saw AIs take one Officers attitudes and use it to shut down security as a whole.

[Cik]

just remove secborgs tbh fam

this sounds like a good idea until you realize that all of the other borgs are obligated to stop the harm and come charging in with n2o canisters and flood your entire department because they don't have a stunbaton and have no way of stopping sec because of the sechud flash immunity

[Luke Cox]

Based on the idea of "harm though inaction," I would argue that impeding security as an AI when there are known antags on the station should be considered harm.

just remove secborgs tbh fam

Borgs are just a nuisance. The AI bolting everything is what I think should be bannable.

[Wyzack]

It really pisses me off when borgs ignore cries of suffering and conflict in the station to follow me around the brig and babysit a prisoner so he doesn't get hurt

[Zilenan91]

Nah, the sheer existence of secborgs makes AIs way way way too effective at enforcing their laws.

[onleavedontatme]

Bolting down a single security officer inevitably leads to the rest of them trying to kill you.

Maybe don't execute people in front of an asimov AI. May as well just make the AI a security officer if it isnt allowed to protect antags anymore.

[Luke Cox]

Bolting down a single security officer inevitably leads to the rest of them trying to kill you.

Maybe don't execute people in front of an asimov AI. May as well just make the AI a security officer if it isnt allowed to protect antags anymore.

I'll drag prisoners into the "prisoner transfer room" and the AI will send borgs to see what's happening. It's blatant powergaming and the admins never do anything about it.

[Cik]

i mean, are you ahelping it when it happens?

that isn't actually supposed to be allowed without prior discovery earlier in THAT round.

at least, last time i checked.

[killerx09]

We could cut the wires on the AI control wires at to the door to the execution room. The reason they're named like Prisoner Transfer/Prisoner Education rooms and lack cameras is to fool an Asimov AI.

[Screemonster]

We could cut the wires on the AI control wires at to the door to the execution room. The reason they're named like Prisoner Transfer/Prisoner Education rooms and lack cameras is to fool an Asimov AI.

Is it possible to map that in or would the random wires each round fuck it up?

[Cik]

i think it may be possible;

ideally though there should be a manual door (similar to the atmos manual valves) which can only be opened by creatures with hands.

bolt control for the AI is optional.

[newfren]

Bolting down a single security officer inevitably leads to the rest of them trying to kill you.

Maybe don't execute people in front of an asimov AI. May as well just make the AI a security officer if it isnt allowed to protect antags anymore.

I'll drag prisoners into the "prisoner transfer room" and the AI will send borgs to see what's happening. It's blatant powergaming and the admins never do anything about it.

If a borg sees you drag a guy into a room the AI doesn't have sight to he's allowed to follow you and ask you what you're doing. It just seems... sensible?

[Steelpoint]

I did initally have it that all doors to and in the prisoner transfer centre had all AI control wires cut so that the AI could not cut if off, it is very easy to do.

However I think someone may have accidently reverted it during the one of many remappings of the transfer centre.

[Screemonster]

i mean, are you ahelping it when it happens?

that isn't actually supposed to be allowed without prior discovery earlier in THAT round.

at least, last time i checked.

Having the same policy that non-antags have for antags would probably be sensible. ie. you can bolt it down if you discover it, but you're not allowed to go LOOKING for it without prior cause.

[onleavedontatme]

Bolting down a single security officer inevitably leads to the rest of them trying to kill you.

Maybe don't execute people in front of an asimov AI. May as well just make the AI a security officer if it isnt allowed to protect antags anymore.

I'll drag prisoners into the "prisoner transfer room" and the AI will send borgs to see what's happening. It's blatant powergaming and the admins never do anything about it.

>not letting security murder people is powergaming

The word has lost all meaning

[TheNightingale]

If I hear the HoS say "I'm going to execute this guy in the prisoner transfer room", WEEOOOWEEOOOWEEOOOWEEOOO harm alarm, send in the borgs.
If I hear the HoS say "I'm going to execute this guy", WEEOOOWEEOOOWEEOOOWEEOOO harm alarm, send in the borgs.
If I hear the HoS say "I'm going to take this guy to the prisoner transfer room", I'll let them, and say something along the lines of "Sure, but can you fix my camera while you're there? I don't have coverage in the transfer room".

Without evidence or reasonable belief that people are being harmed, you can't assume it. And remember, whilst one officer might be a rogue cop (so bolt them down, get Beepsky to arrest them, and have another Security member bring them in), if the whole team is working together to harm people, that's not okay. If you see someone harming a prisoner for a good reason (they're shouting "SEC ROGUE", for example), get the officer in question to promise not to do it again. Whilst all harm is bad, your response can change so long as it stops people being hurt.

[Scott]

If an asimov silicon sees you killing people you are perfectly valid for some asimov dicking.

With that said, don't turn checking out the transfer room into policy, good silicon players will accept it for what it appears to be until situations lead them to enter it and "become aware of the truth".

[Jazaen]

The good way to kill prisoners is to sign them into perma first, then make sure intercoms are down and get them into the transfer room. If AI asks where they went, tell it they escaped (or are braindead at the bathroom, or were sent to gulag that conveniently had some of it's cameras cut by someone beforehand (act like they were damaged all the time) ).

[Steelpoint]

So long as Silicon's at least give Security the initial benefit of the doubt.

The very definition of the transfer room is the transfer of people, it should not seem suspicious that people who get sent into it don't leave considering the 'purpose' of the room.

I think one issue is random crew members would spout how the room is just a execution room, which gives Silicon's enough of a excuse to investigate it.

[Zilenan91]

Well yeah that's because we all know it's the execution room, not the transfer room

[Saegrimr]

Yeah if you've got a group of officers beating down prisoners in the main hall, you earned it.

Steelpoint, Nightingale, and Cik got the right ideas here.

If you're a silicon you shouldn't care about a room you can't see labeled "Transfer" unless some idiot is yelling "HEE HAW TIME FOR EXECUTIONS" and dragging them there.

[Cheimon]

Yeah, just confiscate their radio while you're searching them (checking for stolen encryption keys, boss), take them up to perma, and if you need to ask permission to transfer them. Everyone knows what "transfer" means but a good AI will give you the benefit of the doubt (oh, we have this surgery kit, so we can change their identities, it's a witness protection thing).

Frankly, if you're needing to use lethal force a lot more you NEED to think about subverting the AI. They are not meant to be asimov forever. There is a whole room set up just so you can do this: all you need is a captain that's not got blood all over him and someone else to keep the turrets off. If the AI's being a huge bastard about letting you in, so long as you have the captain onside he has an RCD and all the access he needs to fiddle with things once there, unless the AI is genuinely rogue it can't do a lot to stop you.

[Malkevin]

I get that Asimov AIs have to try to prevent executions and harm to prisoners, but for the love of god can we please make it a rule 1 violation to bolt in every single security officer and cripple sec if the AI decides that somebody in sec caused harm?

Already a rule:
https://tgstation13.org/wiki/Rules#Secu ... d_Silicons

Security and Silicons
Silicons may choose whether to follow or enforce Space Law from moment to moment unless on a relevant lawset and/or given relevant orders.
Enforcement of space law, when chosen to be done, must still answer to server rules and all laws before Space Law.
Silicons are not given any pre-shift orders from CentCom to uphold access levels, Space Law, etc.
Releasing prisoners, locking down security without likely future harm, or otherwise sabotaging the security team when not obligated to by laws is a violation of Server Rule 1. Act in good faith.
Intentionally acting without adequate information about security situations, particularly to hinder security, is a violation of Server Rule 1.
Nonviolent prisoners cannot be assumed harmful and violent prisoners cannot be assumed nonharmful. If you do not know the nature of their crime, see 1.3.2.1 for details.
Releasing a harmful criminal is a harmful act.

[Cik]

Yeah, just confiscate their radio while you're searching them (checking for stolen encryption keys, boss), take them up to perma, and if you need to ask permission to transfer them. Everyone knows what "transfer" means but a good AI will give you the benefit of the doubt (oh, we have this surgery kit, so we can change their identities, it's a witness protection thing).

Frankly, if you're needing to use lethal force a lot more you NEED to think about subverting the AI. They are not meant to be asimov forever. There is a whole room set up just so you can do this: all you need is a captain that's not got blood all over him and someone else to keep the turrets off. If the AI's being a huge bastard about letting you in, so long as you have the captain onside he has an RCD and all the access he needs to fiddle with things once there, unless the AI is genuinely rogue it can't do a lot to stop you.

yeah this too

people (especially HOS and captains(!)) complain so often about asimov AIs cockblocking their rampages, but all they had to do was spend 30 seconds in the upload and it would have been no issue.

if you are playing sec or head in gang or rev, there is NO EXCUSE not to subvert the AI to be friendly to your cause (unless you just want to handicap yourself)

i get a ton of grief about asimov when playing AI but protip there is a reason i don't bolt the upload.

[Malkevin]

Bolting down a single security officer inevitably leads to the rest of them trying to kill you.

Maybe don't execute people in front of an asimov AI. May as well just make the AI a security officer if it isnt allowed to protect antags anymore.

I'll drag prisoners into the "prisoner transfer room" and the AI will send borgs to see what's happening. It's blatant powergaming and the admins never do anything about it.

>not letting security murder people is powergaming

The word has lost all meaning

Don't be daft Kor.

You know full well that the reason the execution chamber is called the transfer chamber, and why the original execution chamber was kept off cams, was so that lore-wise asimovs wouldn't know about it and have no reason to investigate prisoners dragged there and OOCly allowed sec to do their job without some nerd with a god complex sperging over security because they finally found an excuse to dominate the other authoritarians so they can feel like the chief authority on station.


Still, if you're dragging people off to the area that everyone OOCly knows is an execution room and serves no other purpose, or worse, openly talking about killing the prisoner you're a fucking idiot and deserve everything you get.
The gulag is the perfect place to drag people off to off them. Away from station so no one ever stumbles across it, the AI has no eyes there, you have a legitimate, non-harmful, reason for carting capital criminals to there, and most of the prisoners kill themselves so no one will be concerned that the criminal never came back or that their suit sensors say they're dead.

[Jazaen]

Pretty sure AI has cameras on the gulag. Just silently disable those, and then cause a workplace accident.

[MrMindless]

I was the AI during the round that is being largely discussed in this thread and I believe some things deserve to be cleared up a bit.
While I do not know entirely how the conflict started and I did not order the cyborg in question to observe the prisoner transfer center the moment I did get involved, I directly saw the results of an execution with the security cyborg informing me an execution did take place.
At this point the head of security, and may I add ONLY the head of security, was locked and separated, at which point he immediately started yelling Rogue Cyborgs (understandable) and making false claims about attempted murder by the cyborg (less understandable).
Eventually the head of security was released from his solitary confinement and two other security members were present, at which point I requested they take the head of security in custody. However the officer present stunned the security cyborg, released the head of security and destroyed the cyborg. At this point I did indeed lock the 3 of them in perma brig.
Eventually the warden, who I must say had a much better grasp on the asimov lawset threatened to murder the head of security if not released from perma brig, at which point the warden and the officer were released. The head of security was not near the door and I closed it quickly after the warden and officer had left leaving him bolted in once again reducing the number of officers related to the murder locked in to the absolute minimum. I did later on attempt to lock down the officer who destroyed my cyborg as well however this was a temporary measure.
At this point the head of security kept yelling about rogue AI and refused any constructive dialog so he remained locked in until released by an officer with a toolbox handy.
After which he continued executing inmates, this time wary of using the perma brig doing so in plain view.

Also useful to note is that there were other officers within security who at no point were locked in because they simply were not involved in any murders.

Locking down all of security as AI is indeed something that should not happen but I believe when a member of security kills an inmate then immediately antagonizes the silicons trying to enforce their laws, they earned repercussions.

[Luke Cox]

The main issue here isn't the fact that AIs are trying to prevent harm to prisoners, it's the fact that they're gimping an entire department because of the actions of one or two people and causing even more harm in the process. Nine times out of ten the AIs in question are more concerned with griefing sec than they are actually stopping the harm. In the latest round, the sec borg let the guy who I gave a lethal injection to die while he detained sec, and never took him to medbay. Additionally, there is the fact that borgs and AIs often powergame the whole "prisoner transfer room as an execution room" part.

[MrMindless]

Powergaming is an issue in every role, head of security being a fairly prominent one as well. Perhaps the cyborg did go out of line checking prisoner transfer however once it happened it should have been your responsibility as well to de-escalate the situation, not go killing the cyborg and screaming AI rogue.

[Luke Cox]

Powergaming is an issue in every role, head of security being a fairly prominent one as well. Perhaps the cyborg did go out of line checking prisoner transfer however once it happened it should have been your responsibility as well to de-escalate the situation, not go killing the cyborg and screaming AI rogue.

Well, bucklecuffing me in the transfer room, leaving a human to die, and later trying to suffocate me with the transfer room vents makes you look pretty fucking rogue

[lumipharon]

The sec officers in that story caused no human harm. They killed a borg, not a human.

You are obliged to follow law 2 orders (ie: let me out of perma AI), unless you have reasonable reason to believe it will result in human harm.
Killing the borg for following it's laws is very shitty (and in fact against the rules, but it depends on context and I wasn't there), but it doesn't let you ignore your laws and keep them bolted down.

That said, if you saw the HoS kill someone and was very confident he would do so again if let out, then yeah, it would be ok to keep him locked down, possibly with the sec officers being unfortunately stuck with him.

That's the key though. Once human harm has happened, that's tough shit, but ultimately you don't actually care about past harm. You care about present and probable future harm.
Killing someone doesn't (always) mean they're going to do it again.

A guy running down the halls with an e-sword covered in blood is most likely going to harm the fuck out of people.

The sec/captain that just executed a wizard are however far less likely to decide to start shooting up other people.
So while you'd do your best to stop the wizard's execution in the first place, I wouldn't keep them locked down afterwards, since that's not going to stop any further harm.

So again, I can't really judge the round people are talking about since I wasn't there, but the general rule of thumb I use is to first bolt the sec officer that just killed someone down, ask them what the fuck they're doing and why (ie: could have been a changling, self defence etc etc), then if they have a reasonable excuse, (such as the previous examples) or they can reasonably convince you that it won't happen again (executed the lone super powered traitor as opposed to killing the first of many cultists, for example), then you can let them go.
If they're saying 'fuck you AI I'm the motherfucking law, I do what I want", or they're clearly just going to keep on executing cultists etc/etc, then yeah, keep them locked down as best you can.

I would only bolt down ALL of sec if they're basically just going on a fucking rampage and turbomurdering loads of people.

[Lumbermancer]

That's a good discussion to have. What should AI do when security murders people? Because there's not much it can do, especially when not having a borg, but to try to lock down harm perpetrators. I also revoke the law 2 privelage to them, albeit I don't even know if it should work within the asimov lawset.

I had a round, I think it was gang, where Warden gunned down a gangbanger in his office. So I bolted him down. HoS was in the office too. Well they of course broke out, arguing that I should open because there's no harm anymore and insulting me etc. What should I do next then? Should I bolt whole sec to lock down Warden? I knew IC, Warden will kill again. Well I didn't. Warden got mad and said something about killing me, so I threw "come at me bro" at him. So later in the round, I hear welding, I jump to core and I see Warden breaking in. With the intent to kill me. For following my laws. Meanwhile Captain and others were breaking into upload with the intent to Robocop me, which I denied access to per Asimov Law 1. I was busy with the Warden, they managed to force robocop me, just as Warden started to destroy my core turrets. I'm confident if they didn't change my laws, Warden would kill me. Which of course I would ahelp, and I would be right doing so, but it just shows example of what can AI do, and what can happen when it doesn't.

I don't think escalation is the right thing to do like other people say.

at first sight of harm i talk to the individual officer via holopad, if he is available or the HOS. i secure a promise that they will not do it again.

Wow, you secured a promise. And they broke the promise. And consequently you broke Asimov Law 1, through inaction you allowed human to be harmed. Unless it's demonstrated that harm was accidental, or harmed person says its ok, you don't have much leeway to act differently than to deal with perpetrator immediately.

I mean that's how Asimov works in the books, it doesn't really work well at all. It's a meme lawset created to facilitate interesting stories. So you best deal with the consequences, and we don't even follow the lawset literally.

[Cik]

if they harm again the gloves come off.

i generally give a little leeway to sec because it's a hard job already. escalating to mass stealthbolt/depower/closeallblastdoors on the first tiny instance of harm on their part just gets them, you, or the station at large killed.

letting a little bit slide with a warning preserves station function and allows the rest of the crew to not get slaughtered while security is busy with you and while you are busy with security.

[Steelpoint]

I would love to know how the fuck your meant to de-escalate a situation with a Asimov Silicon after they "coincidently" discover your executions of high level prisoners, Gulag or otherwise.

The exact moment they get that switch in their brain that ticks from 0 to 1 is the exact moment they're glued to you like a dog in heat, usually it ends with either myself and the rest of sec being bolted into the brig or that Silicon thrown out a airlock and the AI screaming how all of Sec is rogue, meanwhile the round's antagonists suddenly get free reign to do whatever as anyone with the power to stop them are either too busying fighting Sec or trying to get out of the Brig and do their job.

[TheNightingale]

If you're caught executing one person as Security (say, someone who killed lots of people), try an apology and a promise not to do it again. (If you have to do it again, don't get caught.)

If it's a gang round and you're killing all the lieutenants and bosses you find, don't be surprised when you're locked down and the bosses are hidden safely away. Try exiling them to the gateway (just throw them through with an exile implant, problem solved).

[Steelpoint]

That's the problem, apologizing is often not enough for many AI's I run into, sure I might be biased and simply forgetting the times when the AI is more lenient, but due to the power of Security and the HoS role (in theory) its really hard to convince the Silicon's I won't execute someone again.

I can try doing the deed on the Gulag but to do that I have to pre-empt by going there and cutting a few camera's, and boy let me tell you that even then on some occasions I see a engi-borg repair those cameras.

[Luke Cox]

That's the problem, apologizing is often not enough for many AI's I run into, sure I might be biased and simply forgetting the times when the AI is more lenient, but due to the power of Security and the HoS role (in theory) its really hard to convince the Silicon's I won't execute someone again.

I can try doing the deed on the Gulag but to do that I have to pre-empt by going there and cutting a few camera's, and boy let me tell you that even then on some occasions I see a engi-borg repair those cameras.

At this point, I feel like the only solution is to remove all the cameras from security and gulag by default, and issue law 2 orders to stay out of security unless specifically instructed. A lot of you are acting like security flaunts executions in front of the AI. This is not the case. In many cases, the AI decides in advance that they want to fuck with security for shits and giggles, and watch the brig like a hawk, just waiting for an excuse to bolt everything. If the AI is Asimov, I go out of my way to do the execution in the prisoner transfer room and make no mention of it outside of PDA messages. Despite all of this, a significant portion of the time, an AI will either watch the prison wing constantly if traitors are there, or send borgs into the transfer room for some bullshit reason ("there was an air alarm!"), fully intending to find a reason to shut down sec.

If sec kills people in front of an Asimov AI, that's one thing. If an AI goes out of their way to fuck with sec, jobbans need to be handed out.

[onleavedontatme]

You know you are not actually obligated to immediatley murder people you arrest. It isn't actually gameplay critical in most cases.

AIs pretending to not hear/notice prisoners screaming they're being executed is far more prevelant than what the OP is describing anyway.

Seriously the AI laws are designed this way on purpose, this is a feature not a bug.

>BUT HOW DOES THE AI KNOW

The average assistant knows the inner workings of every piece of syndicate technology, not a huge fucking leap for the super computer to understand that people committing capital crimes might be subject to execution.

[Luke Cox]

You know you are not actually obligated to immediatley murder people you arrest. It isn't actually gameplay critical in most cases.

AIs pretending to not hear/notice prisoners screaming they're being executed is far more prevelant than what the OP is describing anyway.

Seriously the AI laws are designed this way on purpose, this is a feature not a bug.

>BUT HOW DOES THE AI KNOW

The average assistant knows the inner workings of every piece of syndicate technology, not a huge fucking leap for the super computer to understand that people committing capital crimes might be subject to execution.

It's either kill/borg or sit in perma. I'd rather be killed and live on as a borg.

As for the whole powergaming aspect, unless the AI has reason to believe that somebody being taken into a "prisoner transfer room" (and this is true as far as the AI knows) is going to be harmed, they have no business interfering. The whole damn point of calling it that is to give AIs an IC reason to not fuck with executions.

[Malkevin]

I had a round, I think it was gang, where Warden gunned down a gangbanger in his office. So I bolted him down. HoS was in the office too. Well they of course broke out, arguing that I should open because there's no harm anymore and insulting me etc. What should I do next then? Should I bolt whole sec to lock down Warden? I knew IC, Warden will kill again. Well I didn't. Warden got mad and said something about killing me, so I threw "come at me bro" at him. So later in the round, I hear welding, I jump to core and I see Warden breaking in. With the intent to kill me. For following my laws. Meanwhile Captain and others were breaking into upload with the intent to Robocop me, which I denied access to per Asimov Law 1. I was busy with the Warden, they managed to force robocop me, just as Warden started to destroy my core turrets. I'm confident if they didn't change my laws, Warden would kill me. Which of course I would ahelp, and I would be right doing so, but it just shows example of what can AI do, and what can happen when it doesn't.

Technically you shouldn't give two shits about harm that might happen, theres nothing in law 1 that says you need to preemptively stop all harm (don't do harm, don't just sit watch and laugh as the clown chokes outside the airlock - those are elements of law 1) or even for that matter give a shit about past harm (no, law 1 doesn't require you to clone that corpse).

Do not be the cause of harm. (The robot is forbidden from harming a human, the law doesn't say anything about forbidding non-robots harming people)
Do not allow harm to occur in the present (that is, something that definitely will cause harm in the now, such as: traitor on a murder boner, sec beating a prisoner to death, rogue mulebot out for blood, a roaring fire, etc), and the literal meaning is you only care about stopping that human coming to harm, whats causing that harm is completely irrelevant.

Anything beyond that is going beyond your laws, and you shouldn't do it.

[lumipharon]

A big thing is simply playing in a way to make the game enjoyable for people.

Like, this works on boths sides of the spectrum - AI's that let sec get away with murder, and AI's that cockblock sec hard and make their job a fucking nightmare because of some harm.

Yes you have laws to follow, but it's a game. Follow them in a way that makes the game enjoyable.
Locking down everyone because of harm that has already happened, when that action is not directly stopping more harm (and usually just results in smashed windows and sec getting really shitty with the AI) is not fun for anyone, and is a shitty way to interpret your laws.

[Cheimon]

To all those arguing that AIs don't care about past harm, remember that silicon policy specifically states "release of a harmful criminal is a harmful act". An unauthorised execution is a murder; those who murder are harmful criminals.

So subvert the AI in advance. It's not difficult: even something along the lines of "4. Legal executions by Captain [Name] and HOS [Name] are not harmful." will do fine. Security doesn't normally need to murder, and in the few rounds it does you can subvert the AI to let you do that once you find out it's going to be one of those.

And remember, if you're killing people to borg them on the assumption they'd prefer that, you DON'T have to guess! You can just ask them, and they'll normally say yes or no. Then, you can debrain them in transfer.

[Malkevin]

To all those arguing that AIs don't care about past harm, remember that silicon policy specifically states "release of a harmful criminal is a harmful act". An unauthorised execution is a murder; those who murder are harmful criminals.

There's laws and then there's rules
The rules are there to stop ais being dicks, they bare no weight on the interpretation of the law beyond what restrictions the rules imply

[Steelpoint]

Imo the only time a rogue/obnoxious AI bolting everything down becomes a problem is if I'm currently in the perma wing at the time they decide to do so. Otherwise with the usage of a Laser or Stun Baton its possible for you to break and gain access to the entire birg from any point, and you can then employ the usage of barricades to act as a makeshift door.

I̶n̶ ̶p̶e̶r̶m̶a̶ ̶i̶t̶s̶ ̶p̶o̶s̶s̶i̶b̶l̶e̶ ̶t̶o̶ ̶s̶m̶a̶s̶h̶ ̶t̶h̶r̶o̶u̶g̶h̶ ̶t̶h̶e̶ ̶g̶l̶a̶s̶s̶ ̶a̶n̶d̶ ̶s̶p̶a̶c̶e̶ ̶w̶a̶l̶k̶ ̶t̶o̶ ̶a̶n̶o̶t̶h̶e̶r̶ ̶w̶i̶n̶d̶o̶w̶,̶ ̶b̶u̶t̶ ̶t̶h̶a̶t̶ ̶r̶e̶l̶i̶e̶s̶ ̶o̶n̶ ̶s̶o̶m̶e̶ ̶l̶u̶c̶k̶ ̶a̶n̶d̶ ̶t̶w̶o̶ ̶l̶a̶s̶e̶r̶ ̶g̶u̶n̶s̶,̶ ̶g̶i̶v̶e̶ ̶o̶r̶ ̶t̶a̶k̶e̶.̶ ̶A̶l̶s̶o̶ ̶i̶t̶'̶l̶l̶ ̶r̶e̶n̶d̶e̶r̶ ̶p̶e̶r̶m̶a̶ ̶u̶n̶u̶s̶a̶b̶l̶e̶.̶ ̶S̶a̶d̶l̶y̶ ̶t̶h̶r̶e̶a̶t̶e̶n̶i̶n̶g̶ ̶t̶o̶ ̶b̶r̶e̶a̶k̶ ̶i̶n̶t̶o̶ ̶s̶p̶a̶c̶e̶ ̶d̶o̶e̶s̶ ̶n̶o̶t̶ ̶f̶a̶l̶l̶ ̶u̶n̶d̶e̶r̶ ̶t̶h̶e̶ ̶A̶I̶'̶s̶ ̶l̶a̶w̶s̶ ̶a̶s̶ ̶i̶t̶ ̶c̶a̶n̶ ̶i̶g̶n̶o̶r̶e̶ ̶s̶u̶i̶c̶i̶d̶e̶ ̶s̶o̶ ̶y̶o̶u̶ ̶c̶a̶n̶'̶t̶ ̶l̶e̶v̶e̶r̶a̶g̶e̶ ̶t̶h̶a̶t̶ ̶t̶o̶ ̶g̶e̶t̶ ̶o̶u̶t̶ ̶o̶f̶ ̶t̶h̶e̶ ̶p̶e̶r̶m̶a̶ ̶w̶i̶n̶g̶.̶

Scratch that I keep forgetting there's no accessible window in the perma hallway, unless you have a toolkit then your fucked.

The main problem with Silicon's going nuts is when they throw a Sec Borg at you, otherwise the worst the AI can do is painfully slow you down. So while it may not necessarily be fatal to sec, but it will all but make any job for the round's antagonists much easier to accomplish when for every step Security has to take they have to spend several minutes getting their footing before taking the next step.

[Anonmare]

I give sec a fair amount of leeway (More than I should if I'm being honest with myself) since letting them do their job unimpeded is a good harm preventative measure and I don't bother with the transfer room. As far as I'm concerned, it's a transfer room. If a prisoner says that he's afraid for his life and asks that I follow him to make sure he isn't hurt then I have to follow him, even into the room in question.
But Security can usually get rid of me by making up a story about a Human hurt somewhere and needing help. Like in space which is believable enough to send a borg to do.
Treat shit silicon players as shit players, if they arrest the officer while the prisoner is dying of cyanide then the borg deserves to be ahelped. That said, it might not have noticed or it didn't stop you in time before detaining you so don't ahelp them at the slightest provocation.

Plus security can be infiltrated by antags, and therefore, it's not always trustworthy so one officer's word that no harm will come to a prisoner is no more believable than the HoP's or the Clown's. I know I'd get ahelped to shit and back if I let a Changeling Officer eat a prisoner if I didn't exercise reasonable doubt.