I've been working on a convention scenario (standard 4-5 hour time slot, 5 players) which will take place roughly 75% in cyberspace. The PC:s are five subsystems (infiltration, maintenance, counter-intrusion etc) of an unwitting AI.

Now I just need to decide on a simple rules system for all this electronic interfering, but which one is best? I have some sketches on a very simplified rules system of my own, but I would want to look for the ones already in existence before I write my own content.

The current idea is to let the players assimilate and control various slave systems inside their home vault and outside, limited by their current processing power. Every skill or system under their direct control that doesn't have it's own dedicated hardware takes up some of their attention.

What kind of cyberspace are you going for here? Tron? Gibson? Or something else?
Or to expand the metaphor, is it a dramatisation of the computer hardware/softwear functions, or of mathematical functions inherent in computing (hardcore!) or is it a kind of "too human" cyberspace as magic land?
I suspect not the latter, and I can't think how to make the second approachable, (which is sort of the opposite problem in each case), but I'll need some grounding in the feel of the setting before I can recommend stuff.

As a random thought, you could give players a common resource pool (a la clock cycles), but then you would have to ensure there were bonus-style enabling mechanics to discourage people from getting all antisocial!

I tend to go for hard sci-fi when I can. There will be no racing bikes, kung fu or "game zone" sillyness. The Gibsonian cyberspace (paths of light, black spears of IC) is perhaps the closest I have when it comes to visualising this world for the players. I would like to keep it as close to computing as possible (at 2018 level, with neural-net AI around). The end product is supposed to be simplified. Ideally it should be explainable to five convention visitors in under 15 minutes. They won't know anything in the beginning, but will gradually become aware. I wanted to try writing a scenario for non-corporeal characters, and I'm just more used to computers and AI than spooks and ghosts. What happens when a can of forgotten liquid placed on your main storage unit becomes an overbearing obstacle?

Right now the PCs have a stat for their available processing power. We can call it, uhm, "Processing Units". Continously monitoring the security system might take up one PU, while running a complex search or using a specialised program like the classical Kuang Grade Mark IV might fill up 2-3 PU. I'm still working out the proportions and some sample systems. Skills are at a 1-20 scale, and can be programed using PU as well. The skills will be very basic, maybe 4-8 of them. The "Attack" skill will, for example, cover all attempts at closing access to a system. Human and drone NPCs will have a few physical skills as well, plus one or two unique ones for special NPCs like the Time Traveller or the psychic.

The scenario starts during a minor crisis (Brussels is trying to unsuccessfully patch over the responsibilities of the mysteriously offline London AI to their own systems, with an ensuing breakdown in services), with the PCs unaware of their surroundings. The only feed still (barely) running is a training simulation modeled on a small village. The PCs are subsystems, each in control of one fraction of the bunker. One used to handle physical service and security, one handled data monitoring and so on. Each of them start with roughly the same number of hardwired skills and PU, with the possibility to add or restore other systems. They can pool it together, or try to carve out their own little fiefs in the bunker.

My first thought was that you need to create some kind of landscape, a context for where things are, or people will not be able to keep track of where they can go, and it may feel like some text based adventure game (Exits are Left, Right and Dennis). What do you think of the idea of having people build up that synoptic view themselves, by sending daemons into other hardware etc as their eyes and ears? Presumably as a self aware AI you cannot simply jump into a hard disk, as you will not be able to execute (have 0 PUs?). My thinking is starting to go old school map drawing + double agents, + setting traps on high ground (you can download to another location, to get more PUs, but people could have set a virtualisation trap there that reworks your stuff). But looking at it this explorative/detective approach is more suited to a more long term game. Such a game would need quite a lot of diversity of landscape, to make the core holding down space/fighting the other players  AIs feel worthwhile. Perhaps meeting other AI fragments with useful privileges but dubious motivation, or even humans on remote terminals who 'know" what happened, a nice inversion of the Marathon style situation.

Are you planning to give each player a go at the physical world? In other words, presumably to focus on the incorporerality of the characters you will want a certain amount of corporeality in the rest of the environment and other people, but the current structure of roles seems to give that job primarily to one of the players. If not then perhaps it would be good to work out what is so interesting and different about being an AI compared to a human being. Now the king with agents approach above does some of that, but I'm sure there are other ways.

In terms of the mechanics, it seems like initiative might be different in a computer world, either fighting for specific interrupts that would give people particular power, or perhaps deciding whether to invest PUs in tasks that will take two or three turns to run. To put it another way, you know how in some hacking things people will say "we've got 2 minuets until they find us" etc, what if the anti-intrusion systems were conceptualised like slow moving artillery, so you have to narrow down their points of escape or they will get away. So you get this fleeting intrusion skittering through your systems, and you can't let it rest, so you try to shut down conduits they would be using to get away, then run a trace on all processes running, and then you get some complex task thing where they need to roll 3 successes, and you need to roll 2, so they can keep going until they get what they want or jump out. I'm thinking something a bit like Deus Ex's hacking system, but with random variables to judge instead of a time bar and picture.

I don't know of any pre-existing really good hacking system, I think that once you have a good way of working out trade-offs and stakes to provide drama in this more surreal environment, you may just be able to use 1-10 stats on d20 roll under. Now this can lead to a lot of failing, but done in the right way that could still be interesting, as there is always roll again vs the danger of being too slow. To compare to the disconnect option, your AIs could have the option to drop sections of their network, shutting them down to avoid further contamination. I haven't done the binomial distribution business yet, but I would of thought you would want a kind of "But I'm halfway there" effect, which means more than two successes required.

I have no experience with real-life hacking or programming, but...

I've heard that Shadowrun 4th Edition's system for matrix combat and so on, is a fairly good mix between what you'd expect from cyberpunk-style cybercombat, and real life hacking. At the very least, it's a solid system, and it works well in play. The basic system basically consists of the actors rolling their program rating + skill rating as a dice pool, whereas normal actions are resolved by rolling attribute + skill. Hardware has stats which limit what you can do, how fast you can do it, and how well you can resist other actors attempts at removing your access to a specific node. If you want to run a less extravagant game, you can easily do away with all the Matrix-esque narrative. Also, there's some good stuff in the Shadowrun 4 core book about ghosts in the machine and so on, which might interest you or it might not.

If you're going to do a system where processing power is a resource, wouldn't it make sense for actions to have a max resource allocation (how effective a certain program can be at a task), and then letting players spend less resources than this if they want to run multiple programs or just want to keep system resources on hand in case of an emergency.

Also, this sounds like an awesome idea for a game. If you're doing a writeup, I'd love it if you posted it (because this sounds like something I'd really, really, really like to run some day).

The players won't jump around themselves. AIs have been around for a few years, and require a lot of hardware to operate. Physically, they exist in a server complex deep underneath Brussels. The senses they have consist of the various minor systems (security cameras, non-sentient drones, satellites, Google, Echelon, databases, public records and whatever they manage to grab before I start dropping metal rods on them from space), or try to see if they can trust on the numerous European Federation technicians, command personell and military police that are with them in the bunker. If they have the capacity to spare, they can set up a continous feed from those systems. Distributing tasks to other systems and running programs over a large physical distance is OK. You get a few modifiers in some cases, like the Chinese net who has been almost severed from the web except for a few controlled, ineffectual ports.

PU set a limit to how many things you can run simultanously, or how much you can boost your skills. I don't think I've handled speed and initiative that much though. Generally I see initiative as slightly disruptive, and prefer to just go in order. If I have to do it, I prefer a set initiative for everyone, with a minimum of rolling. Most cyberspace combat is for information and influence in various systems. There are a number of broad skills like Search, Create, Attack, Defence and so on, with some special skills. NPC humans and drones in the physical world have other skills appropriate for physical actions. Programs are very specific actions (commanding around a group of cleaning drones with Create, sending Kuang Grade Mark V crashing into NYSE with Attack), but most are just covered with skills and skill bonuses. I'm working on a system for transfering excess PU into skill levels. These characters are working with code on an entirely different level than humans. Most programs are just incorporated into skills. If one of them wants to play Google, they don't have to write it because it's already in the Search skill.

The players aren't 5 AI, they are 5 fragments of an AI. One of them plays the service subsystem (lots of control over internal physical security and drones), one plays the infiltration subsystem (a special Infiltration skill, and high Search), one the security subsystem (a Detection skill, high Attack skill) and so on. None of them should be able to go off alone as they need eachother's support to function.

A major part of the plot is that their central system has been cut off from the rest of the bunker due to a disaster, and is now looping through a training simulation where they are a bunch of kids exploring a village on their summer ferie. The simulation is rapidly degrading while the techies in the real world are trying to fix it as fast as they can before WW III drops down on them. Unless the players wake up (basically by accepting that their world isn't real, and terminating the simulation) they grow desperate and force a reboot of the system.

Simplicity comes first in this case. I can't assume that everyone at a con will know hacking, Internet architecture and Lagrange points. There is also a limit to how deeply I will be able to study this. I'm just a single amateur, after all. The time limit for cons around here is 4-5 hours.

The "landscape" of the Internet will be the classic Gibsonian paths of light. Physical distances aren't really relevant, as they can pretty much zoom around in nanoseconds. There are other AI around, from the shy number crunching system that handles data for SETI to the very outgoing black market security program, or the Vatican secretary AI that dreams about being a pirate. AIs are expensive, so there won't be one in every coffe machine, but have been around long enough and become cheap enough that most major corporations, data havens, ministries and economic centra have their own and that people are somewhat used to having them around. They are all individuals with their own motivations, resources and perspectives, as neural nets in this world develop continously. The group they will have most contact with is the staff of the bunker, who they might or might not choose to follow. The staff is having their own internal disputes, especially between the civilian technicians and the military command unit. Most of the time humans come in groups. For example a 10-man team of security technicians, an MP squad or a community of 20 amateur hackers. They don't roll individually, but has a combined skill. Then there are a few truly exceptional individuals, like the Time Traveller, the psychic cult leader and others.

Accumulation of victory points sounds great both for tasks and matching two characters. "Setting a new trajectory for the French satellite requires 3 vp, because it's French and therefore way cooler. You have time for 5 rolls, while the ESA security team makes one Detect roll each turn until they find you and can start rolling Attack to boot you out, which you can counter with Defence." And I could use something to measure speed and initiative.

Storywise, the whole thing looks like it's going to be in two parts following a parallell time table (at least until the technicians panic and reboot the entire system). One in the simulation where they are kids, and one in the real world where they are still the same "kids" but in charge of whatever they manage to grab before the nukes hit Brussels. The sim is basically a doll house, a safe environment to "teach" a neural-net AI. Every scene and event inside it is supposed to be a learning experience. From how to deal with your smaller sieblings arguing over saturday morning cartoon shows (diplomacy), visiting the local history museum (information analysis), talking to the old neighbour (social problemsolving) to going at your grandaunt's funeral (death and ageing). All those parts that are hard to set down in hardwritten code, and needs to be evolved.