Three released the magnetic grips that were keeping us attached to the shuttle's hull and said, Initiate in 3-2-1.

It pushed off from the shuttle and we fell through the dark, with a little help from a set of removable EVAC suit-maneuvering thrusters. Three was good at following the instructions from the suit's feed, and we landed lightly on the cargo bot's back, our grips engaging again to keep us in place.

You can just talk, I told Three. I pinged the shuttle to let it know we were in place, then tapped the cargo bot's internal feed and slipped past its wall. Before the bot finished querying me as to what the hell we were doing on its back, I removed its memory of the proximity alert and told it everything was fine, just keep following its schedule. We aren't pretending to be a SecUnit, I added to Three, we're pretending not to be a SecUnit.

The cargo bot hauled itself toward our next target, an entrance to the main cargo sorting port for this dock. Three hesitated for .02 seconds while it parsed what I'd said. It likes using logic. Which not that I don't; it's stupid not to use logic when you're constructing flow charts and databases, etc. But it's not particularly helpful for threat and risk assessment. Priorities reorder too fast in high-threat percentage situations, humans exist, and so on.

(Save-for-later: Check wonky risk assessment module for excessive logic statements. That would explain a lot.) Three finally said, We are pretending to be space debris.

It had a point. And what Three was pretending or not pretending to do wouldn't much matter in the next ten-plus minutes.

The problem with the mission—Okay, the problem with the mission is the mission itself. Missions in general often suck, but extractions that can turn into hostage situations are the worst.

The other worst was that this dock and the section of the torus it serviced was owned by Barish-Estranza and contained one of their second-level headquarters installations. And there was a chance in the high 70s that they knew we were coming.

(Emotion check: Again? It's been like two seconds.)

Obviously, I did not want to do this. But Three was shit at pretending to be human. (I know, I should have been working with it more, "mentoring it," as the humans insist on saying. But we also needed to let Three do what it wanted for a while, and what it wanted was to watch science documentaries and listen to freakishly advanced machine intelligences pretending to be bot-pilots talk at it. (Mostly Holism, but a few of the others had got in on it after discovering that there was a SecUnit around that liked being lectured to.) So there wasn't a lot of pretending-to-be-human training involved in that.) Our cargo bot reached its hatch just as it was cycling open. It climbed inside and Three's dark-vision filter activated, so the airlock's blue and green safety lights lit up more of the interior.

It was a lock designed for bots, so it was pretty utilitarian. There wasn't even any active process to clean off the build-up around the vents. Our suit feed gave us stats on the flush of station air and the pressure and gravity change, and we started picking up ambient audio. Then the inner hatch slid open.

This was a bad moment. We couldn't risk a peek over our cargo bot's shoulder and Three couldn't deploy drones without setting off the dock's interior alarms. Three admitted, This is not ideal.

By which it meant it was somewhat terrifying. I acknowledged but didn't reply; I was waiting to grab the interior feed so we didn't get killed.

(It turns out executing a deliberate, convincing distraction is a lot harder than being a distraction accidentally.)

The cargo bot stepped through the inner hatch and we slipped off its back. Its giant body continued on, even as a flotilla of hauler bots swerved to avoid it. Three had frozen into immobility as I pulled in the dock's feed and started to work. We were in the shelter of the lock's safety overhang, which was in a visual shadow because of the angle of the wall lights; it was also a tiny motion and audio sensor black-out zone. (Since my penetration-testing report on Preservation Station, they had installed individual sensors on all the hatches where that happens. Mihira and New Tideland's stations and space docks were in the process of doing it now.)

(This is why humans shouldn't do their own security, or farm it out to automated sensor sweeps. Sensors can't see what they can't see, right? And while humans are shit at security, mostly because of the overreacting and some being assholes who just like using weapons to hurt each other, they are fantastic at getting into places where they aren't supposed to be.)

(Apparently it's genetic? Overse recommended a book on human evolution and it was even more fucking weird than anything I'd read about it before.)