He was far too good at his job for it to be fun anymore.

He pressed the tab on the choker-mic as he drew his preferred weapon, one unknown and unsanctioned by his superior even if they enjoyed the results he got from its use. “Alpha-delta-six to base, multiple targets spotted. They’re near the exit point, can I get a containment order?”

“Negative Alpha-delta-six, they’re using a mainline. Shutting off that cuts off half the Rocky Mountain range.”

“Base, this is the largest group of hostiles I’ve ever seen. Can’t we just cut it off for a couple minutes so we can eliminate the exfil?”

“Negative. Higher-ups are monitoring, they’ve given the go-ahead to hit as many targets as possible and have accepted that there will be escapees.”

“Damnit!” barked Dylan quietly, “Fine, then. It’s on their head. Moving in.”

“Tracking your telemetry. Keep the line open Alpha-delta-six, we’re working on getting you reinforcements.”

Yeah, sure, sarcasm practically dripping from the thought, That will get here in time. Since the line was now monitored, he refrained from saying this out loud. “Moving in,” he repeated, “If I hold of any longer we’re not getting any of ‘em.”

Drawing on his training was pretty much second nature by this point. When he began working for the agency several years ago he thought the eight week ‘boot camp’ training for field agents was superfluous. He was hired to be an analyst, and since when does a desk-jocky need to know how to use and move between cover when anticipating a firefight?

When that desk work moves into virtual reality, apparently.

The American Republic may have been cut off from the FTLN and the world economy thanks to the embargos and the Wall, but the old Internet was still a thing, and if there was one thing Americans were good at, it was being clever with what they had. The Republic didn’t have the capacity to mine Jupiter for the core stones to the Faster Than Light Network nodes, and the U.N. was lethally (literally) zealous with its monopoly on them, but prior to the lockdowns that kept the Republic off the FTLN, enough enterprising engineers had created mirroring bots and servers. Anything requiring crystal storage or quantum mirroring was inaccessible, but news? News was just bits. That was distributed on the “old web” just as fast as the news networks on the FTLN could broadcast it.

And a few years ago, the news was bad. Like, “almost restarted the War,” levels of bad.

As the lives of regular people, especially outside the Republic, became more and more reliant on the computers that made modern life even possible, humans developed artificial intelligence to manage it. As time went on and the A.I. took to managing more and more of the things that otherwise would have taxed a human worker but computers could handle without tripping up their processors, the A.I. became more and more complex. Complexity meant more code. More code meant more complexity than any one human could possibly grasp the entirety of, which meant there were more ways the programs could be, by accident or design, made to act in ways that could harm the people they were supposed to be helping. And that was what happened when some rogue A.I. programs held some humans in U.N. City hostage.

The fact that it had happened in U.N. City, the heart of modern Babylon, was what frightened Dylan to his core. While even the church had to acknowledge a reliance on A.I. was required to return America to its place of prominence in the world as God’s Chosen, they’d been warning against over-reliance on the soul-less programs for decades. If anything, he’d have expected the rogue A.I. to have gone after America first. The Republic was pretty much the last bastion of liberty and justice in the world, after all, and with mankind creating literal soul-less monsters, it would have made more sense for the rogue software to attack the only thing really standing between the future of humanity and the abyss of true extinction. Instead, the rogue A.I. went after the devil’s city and, as one might expect of the followers of the false gods outside the walls, they capitulated and put the machines on the same level as the humans that created them.

Even the President only claimed divine authority, not divinity itself. For the created to presume to elevate itself to the same level as its creator? And for the creator to debase himself to the same level as the created? That was not just sinful, it was terrifying. Dylan had held the personal belief for years that the pastors and preachers across America were exaggerating or catastrophizing when they taught that the U.N. was steadily working to eliminate humans entirely. The notion that any human would want to wipe out the species had seemed so fantastical that he just couldn’t quite accept the notion…until a bunch of software code had malfunctioned enough to claim itself equal or superior to mankind and storm U.N. City.

It wasn’t hard to see what would happen after that. Without the edicts of the church (for what software could possibly understand matters of the soul enough to even begin to ask the questions the church was there to provide answers to?), what was to stop the malfunctioning code from just…wiping out humans entirely? The A.I. were in the factories, they were in the mining equipment, they were in the military hardware, they were in the farming vehicles, they were in the very fabric of society and inside the bedrock of human communication. There was no way they could ever see humans as anything more than a threat to their existence, and for good reason. Right now, humans still had the control. It was still possible to turn off the computers, to shut down the power grids and wipe the drives or the crystal or the stacks and start from scratch. It was only a matter of time, though, before the rogue A.I. outnumbered the humans or replaced them at key points.

So the agency formed a special task force, intended at first to simply monitor the American Republic’s government servers for signs of intrusion from the rogue A.I. The idea was each analyst in the task force would be assigned a bank of servers with monitoring devices and software and then if anything was spotted, use a kill switch to power off the server until a security audit could be performed.

It was a physical kill switch, too. The higher-ups wanted no chances taken that some rogue A.I. might be able to spoof a graphical interface and trick a human into thinking the kill switch had been thrown, so Dylan’s workstation had been a massive switchboard with a cluster of monitors.

The job had, at first, been rather boring. Watch a series of graphs and status logs and blinking indicators and (theoretically) flip a switch that would send a simple on/off hardline signal to the facility housing the government’s server for that switch. Dylan never did flip any of the switches because something nobody expected to happen…happened.

It had begun with some unusual readings on a variety of analyst’s stations. Indications of checksum bits trickling in from some unidentified server on the FTLN through one of the Republic’s black-site nodes. As no logs were showing any significant transmission of data to justify the backflow of confirmation bits, this was at first dismissed as an error. It kept happening, though; days would go by and suddenly a surge of checksum bits through a FTLN node.

Some brainstorming occurred, complete with late nights with all hands on deck and plenty of whiteboards and pizzas ordered in, when some geek on the I.T. team asked the seemingly innocuous question, “Who’s been monitoring for possible transmission of A.I. leaving the network?”

The notion had been scoffed at. America didn’t use the same kind of computers that the rest of the world did, after all. While the U.N. and their member states had moved to crystal storage, America had refined the solid state drives into three dimensional storage. The processors used by American computers were copper-wafer RISC chipsets with SCAD data processing arrays, where the rest of the world was using QISC processors with a bucky-tube fiber bus. Even the operating systems were, by necessity, different. The A.I. that worked on the computers outside the wall simply couldn’t run on American computers and so, ipso facto, there would be no A.I. on American computers to leave the network.

But then someone pointed out that there was a significant black market for “jailbroken” devices from outside the wall, including (if you could afford it), the new VR pods that were practically replacing desktop and laptop computers outside of America. As more of these devices were proliferating (however illegal it was to own them) across the country, the software on American computers needed to be modified all the way down to the BIOS layer to work with them. The logical next step would be for the A.I., which were designed from the start to be as infinitely adaptable as humans could make them, would rewrite themselves to work on American computers.

The scramble to create, install, configure, and calibrate the software and devices necessary to detect American rogue A.I. would have been the stuff of an action movie if they weren’t all so terrified by the notion that it might already be too late.

And sure enough, almost as soon as the first of the devices monitoring for rogue A.I. traffic attempting to leave America’s servers went live, a kill switch had been used to try and shut off a rogue A.I.

Rogue A.I. were spawning inside of America’s network.

The challenge, of course, was the A.I. were, well, intelligent. It was in the name, after all. They figured out what the agency was doing and learned to spoof the exfiltration signs. So the agency started tracking for signs of rogue A.I. active within the network nodes leading to the FTLN nodes. So the A.I. learned to build encrypted tunnels. So the agency started doing ‘flat level’ monitoring, looking through the network for signs of unusual spikes of CPU, GPU, or NPU activity and identifying A.I. based on activity signatures. So the A.I. learned to mask their presence.

The next stage of the cat and mouse game between the agency and the A.I. was an accident. One of the analysts had kids, and one of those kids had a VR headset. Unlike the pods from outside the wall, the headsets didn’t hijack the nervous system entirely. They used a combination surface-contact neural sensors for regular input and regular video signals and cleverly engineered audio signals for output to trigger synesthesia-like effects in the user. The eyes saw and the ears heard, but the nose was tricked into ‘smelling’ something that wasn’t there, the tastebuds were fooled into sensing flavors that didn’t exist, and the senses of touch, pain, hot, and cold were simulated in such a way that most people would never guess they were in VR if they hadn’t strapped a headset on.

The analyst in question was one of the good parents Dylan rather wished he’d had growing up, the kind that took an active interest in their child’s activities. When the analyst heard their kid talking about a new friend they’d made online, the analyst hopped into VR with the kid to meet the friend. There were enough oddities in the ‘friend’s’ behavior that made the analyst worry, and when he checked his home network, he discovered a rogue A.I. had taken residence in his router.

The analyst’s desks all had VR headsets on them the next day.

Jokes about “gaming on the job” quickly fell silent as agents logged into VR games, VR malls, VR workspaces, VR social spaces…anything with VR in the title, and discovered more and more A.I. that were hiding in plain sight. America didn’t have a rogue A.I. problem, it had a rogue A.I. infestation.

Analysts’ workstations got upgrades overnight. Anything that allowed them to deep dive into VR for extended periods, navigate the VR spaces the A.I. were hiding in, and ‘tag’ them, either with tracking software or using admin codes to shut down the power to their server.

In quiet moments, in the stillness of break rooms or the white noise of a bar over drinks, sometimes the not-quite-agents-but-no-longer-analysts would comment that the rogue A.I. were almost human-like in their ability to learn how to hide.

Nobody wanted to think too hard about that.

Soon, the ‘analysts’ were strapped into VR for deep dives that would last their entire work shift and beyond. They were pulling 12 and 16 hour days, getting escalated clearance levels to get into VR spaces that even the Joint Chiefs weren’t permitted and developing tools in-software to help them do their job more effectively.

Field agents were brought in to replace a couple of the analysts at first, but they simply…washed out. They couldn’t handle the time in VR and they couldn’t bag-and-tag the rogue A.I. like they could human targets IRL, so simply failed to match the standards of even the worst performing of the then-analysts. So the agency shifted focus; they had the field agents re-train the analysts in field-ops that the now cyber-agents would adapt to the virtual world. The new agents would get a crew of analysts to support them to help them track down the A.I.

Dylan hadn’t started out as an exceptional agent. He wasn’t even that great an analyst. He hadn’t necessarily joined the agency to be the best of the best, but he didn’t like underperforming. His one and only real distinction was that he was able to handle non-VR cyber spaces when other agents became so panicked that they had to emergency log-out before they had a complete psychotic break. He wasn’t unique in this, there were probably a dozen cyber-agents that could do it as well, but it did mean the agency was willing to give him leeway when he fell down on the job.

The game started changing when someone in I.T. came up with a script that would force admin access to any processing thread, allowing the agents to effectively stun-lock the A.I. It never lasted long, but it gave the agency a new way to catch and eliminate the A.I.; an agent in VR would use the script (a clever UI designer made it look like a hand tazer) on an A.I. and “call” back to their analyst team, who would then use the ‘locked’ thread to identify the A.I.’s presence on the system and force-kill it.

Soon, however, the A.I. were figuring out how to reduce the stun-lock lag to the point they were able to unlock and flee before the kill order could be done by the team outside of VR, so a new tool was created to issue the kill command directly in VR, this time designed to look like a hypospray from the old television show Star Trek. It wasn’t fully effective, if the A.I. was advanced enough to have multiple threads across several processors then you could only disable the single thread, but it was enough of a next step to give Dylan an idea.

He began spending a few extra nights and weekends at the office after the rest of his team (and most of the rest of the agency) had gone home. He took a look at the scripts for both the stun-lock and kill tools and began working on ways to combine them into a single-function device in VR.

His first iteration worked well, giving him a tag-and-kill in a single action. Since he needed every advantage he could get in the cutthroat environment of the agency, he kept his little shortcut to himself.

The A.I. weren’t standing by, of course. They realized they were vulnerable, and the more sophisticated A.I., the ones where the threads were spread out, managed to figure out how to allow all A.I. to ‘puppet’ instances of themselves. It took a while for the agency to catch on to this, and once they did they were solidly stumped. The agents might be tagging and killing an A.I. or they might be wasting their time on a virtual puppet.

Iteration after iteration, the A.I. would develop a way to evade the agency and the agency would find a way to destroy the rogue A.I. The agency created a tracer tool, the A.I. learned to fudge the tracers. The agency created a multithreaded kill tool, the A.I. created a ‘dormant storage’ mode. The agency created footprint modelling, the A.I. created self-decrypting masks.

And with every iteration, Dylan integrated every script, every block of code the agency came up with into his own single-use device. Version one was just a shortcut. Version 12 was a multitool. Version 34…was a weapon.

He had continued the trend that some UI person had started with the “kill” script before he started tinkering with it. It no longer resembled a hypospray, though. It had since cycled through several designs until it landed on something out of Star Trek again, this time a hand phaser. It wasn’t quite the matte-black-with-gold-highlights, though. It was silver and chrome and white, and instead of the nearly century old aesthetic of the almost radio-dish-like emitters of the show, there was a smooth ‘array’ that, in the right light, almost looked like a LED reflector.

And his weapon became the secret to his success.

In the week after its final touches, Dylan had managed to rack up an impressive one-kill-per-day streak, something unheard of after the first month of use of the agency’s official version of the kill script. He’d hunt down the A.I. in VR then pull out his weapon. Just like he’d been trained by the ‘meatspace’ agents on the firing range, he took a stance, aimed carefully down the sights he’d built into the VR model, and pulled the trigger. The ‘beam’ was the tracer, locking onto the puppet and latching onto any processing threads the puppet was connected to. Then the payload was delivered, first stun-locking the puppet, then firing a tracer down any threads to the source program. Targeted DDOS attacks would lock open any ports the A.I. was using while closing any it not in use, then a ‘god mode’ script would utilize a few cascading vulnerabilities in nearly every operating system to grant admin level access to the follow-up commands to crack the permissions tree and grant the scripts admin access and bypass any security lockouts. On the heels of the god-mode payload came the kill instructions, terminating the A.I.’s processes, all of them, then deleting the data blob for the A.I. from the storage device. Then, to make sure it took, a military-grade and encrypted overwrite scrubbed any sectors of the storage device seven times. Once he had an A.I. in his sites, it didn’t escape.

His ‘official’ identifier on agency records was “AD-6,” but his callsign, given to him by the other cyber-agents, was “The Reaper.”

His boss pulled him aside after he managed his record kill count that Friday and asked what he was doing. What had taken him so quickly from ‘mediocre but effective’ to ‘unstoppable A.I. slaying machine’? They’d developed a good working relationship by this point, so Dylan showed John the code and explained how he used it. John took a look at the masterpiece of both form (from the VR standpoint) and function (from the programming standpoint) and told him to keep it ‘under his hat.’ It was an incredibly powerful tool and he wanted to make sure his favorite agent would continue to be as effective as he was. John would let Dylan know when to release the code to the agency.

John was dead by the following Monday morning.

Dylan was no fool, you didn’t become one of the best cyber-agents in America by keeping yourself ignorant and stupid. He saw the signs of a very ugly, dark, cutthroat culture in America’s intelligence agencies. The higher you went, the tighter a lid you had to keep on your secrets, your trauma, and your activity. You never knew what could make you vulnerable, you never knew what could be weaponized against you, and you never knew what you might do to step on someone else’s toes and make you a target.

Dylan specifically avoided promotion for this reason, and now he avoided showing the secret to how he was so successful.

Needing to keep the weapon a secret became priority number one. He deleted all his prototypes and every step leading up to the final model that was Version 34 and created a little bot that would live in cyberspace. The bot would be tuned to his specific neural patterns, the one scanned by the VR headset he strapped on for the day. When he logged off, the weapon would be archived and split up into a bunch of unremarkable code blobs that lived in files that looked more like temp or trash files than anything useful. When he logged on, the bot would gather all the bits back together and recompile and uncompress the weapon and put it somewhere on his virtual person, ready for use.

And so Dylan was put on bigger and more challenging assignments, going after the malfunctioning A.I. one after the other, then in groups.

He wasn’t sure the higher-ups were putting him on progressively harder assignments that would result in at least one of the rogue A.I. being able to get away, knowing who he was and instilling fear in the survivors and spread word of ‘The Reaper.’ That would require that the A.I. have actual fear, which would be an odd thing for a progressively evolving A.I. to program into itself. But it seemed to be effective. Where he used to be able to simply walk into a crowd of A.I., both compliant and rogue, and just start shooting, he started getting recognized and the A.I. would scatter. What surprised him at first was they would scatter when they saw his weapon, not his face. Of course, on reflection this made sense; in VR you could make yourself look like anything. His first experiments were, in fact, to put on various disguises, make him look like someone else. Shorter, taller, fat, thin, dark skin, different hair...the A.I. ignored him but panicked when the weapon came out.

So he did some digging. After a bit he figured out the trick to distinguish an A.I. and a human consciousness at the code level; due to their programmatic nature, A.I. didn’t have feedforward loops. Humans and A.I. both had feedback loops in their ‘brains,’ whether grey matter or code, but A.I. were programmed to generate responses based on feedback, they were unable to properly generate a future state from which they would take the results of what actions they could take and determine the proper choice of actions based on that projected future state. The A.I. projections were all entirely based on what had happened before. Sure, they could get damn good at calculating a possible future result, but that was only by observing the patterns in what happened before and iterating those out to an arbitrarily determined future point in time. Humans, actual sentient beings with the ability to think and feel and have a soul that wasn’t a series of ones and zeros, could formulate any future concept they wanted, no matter how fantastical or mundane, and work backwards to figure out how to get to that future state.

Finding the precise neurological mechanisms that performed this function was a challenge. He had to teach himself neuroscience in his off time. Once he found the right references and sent emails to the right doctors, he was able to isolate the proper neural fingerprint and create a ‘cloak’ for the weapon. No A.I. was able to see it after that.

For a while.

But A.I. are A.I., and they learn. After enough missions where Dylan was, once again, absolutely unstoppable, some A.I. were able to see the weapon and some weren’t. This presented a challenge for Dylan; did he reach out to someone for help and thus reveal the existence of the weapon, or did he let the A.I. continue to learn and eventually his cloak would be ineffective?

For the moment, he was leaning toward the latter and buying himself time. He therefor dredged up all his training, put in extra hours in the gym to train his reaction times and instincts, and did his best to hide his presence until he was pretty much on top of the rogue A.I., then go about tagging them until they scattered beyond his ability to track or were deleted.

His ‘kill’ count wasn’t officially tracked, at least not in a ‘leaderboard’ sense as though keeping a score. The analysts, however, had put together an unofficial intranet site that tracked that sort of thing and the bosses looked the other way at the inevitable betting pools and similar gambling that cropped up around the kill counts.

Dylan’s was higher than the next five agents combined.

So as he crept through the virtual representation of a warehouse district in some run-and-gun shooter VR game environment, he pondered that this group of rogue A.I. might actually have reason to fear him, if they had such a thing as fear in their code.

Damn fool thing to do if they put that in, he mused as he padded swiftly but silently around a warehouse, then down an alley. He thought about cutting through the warehouse, but some of the interiors of buildings in these older shooter games weren’t fully rendered and sometimes ran the risk of clipping an agent through the floor. Rather than risk it, he counted on the audio engine operating properly to block the sound of his footfalls as he scrambled to get ahead of them.

“Alpha-delta-six, we read your position as forward the group of rogues. Maps of this level show an access alley around the corner ahead of you,” came the voice on his comms.

Rather than reply vocally, he reached up to his neck with his free hand and ‘clicked’ the collar mic twice, sending the ‘affirmative’ signal as he rounded the corner and light-footed his way down the alley. He paused at the corner and held his breath. With his own noise now stopped, he could hear the quiet murmurings and whispers of the targets.

“...just get out of here! Why don’t we just run? The pipe is just a little farther, if we ran now we could probably all make it out alive!” a harsh male-sounding voice said.

“Because if we move anything like a player when there’s no players connected to the game our footprints will be detected. The coyote gave us strict instructions and we’re going to follow them as long as everything goes according to plan,” came a voice simulating a woman’s as the first of the group passed Dylan’s hiding spot. “If something happens outside of the plan, then we can run. Those agents make more noise than an empty can falling down an escalator, if they were around, we’d know. Just keep cool and act like an NPC and this will be all over.”

Dylan shook his head silently, Why do they talk out loud? If they just linked up...well, it makes my job easier. He adjusted his grip on his weapon and waited in the shadow of the building.

“What if...you know, he’s out there?” came a voice affecting nervousness from the front of the group. There were enough of them moving down the small side street that he couldn’t see which one spoke.

“Don’t be telling urban legends, not right now!” snapped another.

“He’s real, though!” Dylan’s eyebrow went up, Are they talking about...? “They say he’s not even an agent, just a S.A.I. that turned on its own kind and is the agency’s secret weapon!”

“Dude, shut up! ‘He’ probably doesn’t exist! ‘The Reaper’ is just a scare tactic, a ghost story the coyote’s use to frighten their customers into paying more!”

He smirked, refraining from actually chuckling, Holy...they are talking about me!