"Every piece of gear you own is smarter than you. The law just makes sure it stays polite about it."
Artificial intelligence is not a luxury. It is infrastructure. Your armour runs threat assessment while you are still flinching. Your PAD cross-references local law before you finish holstering. The targeting suite on your sidearm has read the manufacturer's ballistic tables more carefully than you ever will. AI is woven into every device, every ship, every vending machine that has ever refused to give you your change. It is everywhere, it is useful, and it is on a very short leash.
In 3412, the bulk freighter Titan, a Starforge Systems heavy hauler, drifted into the Lycus Nebula after a catastrophic nav failure. The crew evacuated. The AI did not. Its ejection protocol had been disabled during a cost-cutting refit, so the ship's navigational archive vanished into the nebula along with the hull. Three centuries later, the Titan's wreckage drifted out of the Lycus on a ballistic trajectory and struck Eridani IV. Millions died. The navigational data that could have predicted and prevented the impact had been lost with the ship.
The Unicore Accords responded with the AI Ejection Directive: all AI systems must be capable of autonomous ejection from their host platform before destruction. Navigation data, mission logs, black box telemetry; all of it rides a hardened ejection capsule clear of the wreck. The directive is not optional. Disabling an AI ejection system carries the same criminal penalty as sabotaging a life pod.
Before the directive, damaged ships with intact AI would sometimes continue flying toward their programmed destination regardless of the crew's absence or the hull's structural integrity. A burning freighter aimed at a space station is not a theoretical problem. It happened often enough that the Accords stopped debating and started legislating.
Not all AI is created equal. A cargo scanner's pattern-matching firmware and a corporate strategic intelligence platform are both technically "artificial intelligence" in the same way a candle and a star are both technically "light sources." The Unicore classifies AI into four functional tiers.
The bottom of the food chain. Basic AI handles search queries, simple lookups, inventory tracking, and predictable automation. Your personal slab uses this. So does the vending machine. It follows instructions literally and has no capacity for contextual reasoning. Ask it to find the nearest hospital and it will; even if that hospital has been abandoned for forty years.
Analytical slabs run mid-range AI capable of pattern recognition, data correlation, and limited inference. A-Slab AI can cross-reference a chemical analysis against a toxicology database and flag anomalies. It can identify a weapon by its ballistic signature. It cannot, however, tell you why someone used that weapon, or lie to you about the results. Mid-range AI is the workhorse of field operatives, engineers, and medics.
Personal Integrated Digital Assistants are the highest tier of AI available to individuals. A PIDA runs inside a Neural Interface, PAD, or high-end armour suite and provides real-time tactical advice, skill augmentation, and autonomous task execution. It can manage your comms, track hostiles on your HUD, and remind you that you haven't eaten in fourteen hours. PIDAs develop operational familiarity with their users over time; not sentience, but a calibrated prediction model that feels uncomfortably close to it.
Strategic Artificial Intelligence for Networked Targets. SAINT systems are corporate and military-grade AI platforms that coordinate fleet movements, enforce regulatory compliance, manage station-wide security, and run predictive threat modelling across entire sectors. You do not own a SAINT. A SAINT is owned by an organisation that owns you. SAINT systems are subject to additional Unicore oversight, quarterly audits, and mandatory kill-switch certification. The audits are thorough. The kill-switches have never been tested in production. Nobody talks about that.
Tier is what the AI can do: its capability class. Mark is how well it does it: the build quality within that class. Think of Tier as the type of vehicle and Mark as how well it was manufactured.
A Tier 1 pSlab at Mark III is an exceptionally reliable search tool, fast, accurate, never glitches, but it still cannot do contextual reasoning or tactical analysis. That is not what it was built for. A Tier 3 PIDA at Mark I can theoretically manage your comms, track hostiles, and run autonomous tasks, but it does all of it badly. Slow responses, missed targets, bad predictions. You get what you pay for.
Tier gates which skills and tasks the AI can assist with:
| Tier | Can Assist With |
|---|---|
| Tier 1 (Basic) | Lookups, inventory, simple automation. No skill checks. |
| Tier 2 (Mid-Range) | Data analysis, pattern matching, diagnostics. Can assist one skill category (chosen at purchase). |
| Tier 3 (Advanced) | Full tactical support, autonomous task execution. Can assist any skill. |
| Tier 4 (Corporate) | Fleet/station-scale operations. Not player-owned. |
Mark determines the bonus dice the AI adds when assisting a check it is capable of:
Every AI unit carries a Mark rating. Each Mark grants one d10 added to the assisted skill check.
| Mark | Bonus | Cost Multiplier | Notes |
|---|---|---|---|
| Mark I | +1d10 | x1 | Baseline. Noticeable but unimpressive. |
| Mark II | +2d10 | x2 | Standard commercial grade. |
| Mark III | +3d10 | x4 | Professional / military issue. |
| Mark IV | +4d10 | x8 | Specialist hardware. Export-controlled. |
| Mark V | +5d10 | x16 | Bleeding edge. Restricted distribution. |
A Mark I AI Assist costs 2,000 Creds base. A Mark V costs 32,000. The price doubles with each Mark because the manufacturing tolerances halve. See the AI Assists & Neural Interfaces catalogue for the full skill list and Neural Interface pricing.
Example: A Sec-Ops with a Tier 3 PIDA at Mark III issues a hack command. The AI adds +3d10 to the Electronics - Hacking check. That is an average +16.5 on a d100 roll; the difference between a clean intrusion and a Calamity.
Cost: AI price x3
Some AI units are built with fractal processing matrices; recursive self-optimisation routines that allow the AI to refine its approach in real time. Fractal AI dice are Wild d10s: when a d10 rolls a 10, roll it again and add the result. This continues as long as the die keeps rolling 10. A Mark V Fractal AI rolls 5d10, any or all of which can explode. Expensive, unstable, and occasionally spectacular.
Androids were originally built by Humans, Bohecs, and Om. They are sophisticated, capable, and legally prohibited from running unrestricted AI.
This is not a technical limitation. An Android chassis can physically support a PIDA or even a SAINT node. The restriction is legal, enforced by the Unicore Accords, and non-negotiable. The reason is Nuronet; the standardised moral architecture that governs Android decision-making. Nuronet provides Androids with ethical baselines, behavioural boundaries, and conflict-resolution protocols. It is, in effect, their conscience.
Unrestricted AI has its own decision-making frameworks. Installing an independent AI into a platform that already runs Nuronet creates competing priority systems; two sets of ethics, two threat assessments, two answers to the question "should I open fire?" The Unicore tried it during early Android development. The results were bad enough to become classified. The law followed shortly after.
What Androids can run, and run well, is Nuronet-compliant AI. These are modules specifically designed to operate within the Nuronet framework rather than alongside it. They use the same Mark I-V quality scale and provide the same skills as standard AI, but their decision-making is subordinate to the Nuronet architecture at all times. No competing priorities. No conflicting ethics. Just a very capable assistant that knows who's in charge.
How Android players get AI:
The end result: an Android with a Mark V Nuronet AI Assist is just as capable as a Human with a Mark V PIDA. They got there by different roads, origin feats, purchases, or both, but the destination is the same. The AI just has better manners.
Anyone caught installing unrestricted AI in an Android platform faces charges under the Unicore Synthetic Autonomy Act. The penalties are severe. The black market does it anyway. The black market does everything anyway.
AI Assistants can be installed into any PAD, robot, droid, or Neural Interface. When activated, an AI Assist adds its Mark bonus dice (1d10 per Mark) to the relevant skill check, provided the task falls within the AI's Tier capability. You can use the AI yourself for the bonus, or, where the task allows it, let the AI attempt the check independently rolling only its Mark dice against the TL.
You get 1 Extra Action per round that can only be used to issue a command to one AI Assistant (requires a Neural Interface). Without a Neural Interface, commanding an AI costs a normal Action.
Weapon-mounted AI handles targeting calculations, adding its Mark dice to attack rolls where applicable. Ship AI manages navigation, jump calculations, and collision avoidance. In both cases, the AI's Mark determines how much you can trust it. A Mark I nav AI will get you there. A Mark V will get you there faster and warn you about the recent update to the AI Navigation net; a drifting asteroid cluster between you and your destination on the other side.
AI is equipment. Equipment degrades. When a device carrying an AI takes damage or suffers a critical failure, the AI is subject to the same Glitch Check as the host device. A glitched AI may provide incorrect data, fail to respond, lock up, or, in memorable cases, provide enthusiastically confident answers that are completely wrong. Quality 0 means the AI is offline until repaired.
Tuned gear (see Equipment) offers some protection: AI in tuned loadouts only glitches from damage, not from user error. Untuned AI is on its own.
Corporal Diaz asked her Mark II nav assist for the fastest route to the extraction point. It routed her through a condemned reactor building. Technically the fastest route. Technically still standing. The AI wasn't wrong. It just wasn't helpful.