I worked on A/UX at UniSoft, we got half of the first dev run of MacII boards (I got to find hardware bugs), I stopped attending BMUG and we chose our own internal code name (Pigs in Space), when the project finally leaked it was as "Eagle" Apple's code name (whew!)
Apparently Jobs sold most of his interest in Apple immediately in 1985 in order to finance founding NeXT the same year. Jobs took a number of Apple employees with him to NeXT, but I'm not sure how long he was speculating who he wanted with him. Yada yada yada, Apple and NeXT were in negotiations by 1996. I have heard that A/UX was the first thing Jobs killed when he returned in 1997, but the final release of A/UX was version 3.1.1 in 1995, and Apple abandoned it a year later.
So if it was Steve Jobs they were hiding from, it would have either been very early in the development of A/UX (if that occurred by January 1985) or Jobs somehow started killing A/UX during early NeXT negotiations with Apple in 1996. Or Jobs had some reason to be at Apple during that period due to Pixar(?)
And if it wasn't Steve Jobs, then there was another Steve at Apple that was menacing employees between 1985 and 1996. Steve is a relatively common name, and it would be pretty interesting to learn who that was.
I am a fan of A/UX, though I didn't appreciate being forced to buy a license for it in 1989 by my university's CS department. Though the price dropped by about half by then, along with the MacII hardware, it was a very expensive purchase for a 17yo, about as much as a midrange car at the time.
And yet there's still only one partial source code leak (for System 7.1) publicly available. Microsoft still has them beat on this. ;)
If anything, I would think you'd want to have technical documentation circulated as widely as possible, so application developers don't have to reverse engineer your interfaces (which would take longer and be more error prone).
Also feature parity between systems didn’t exist so they were still trying to one up each other.
Remember that mac 68K system calls were via ("A-line") opcodes, and their only extension/fix mechanism was head- and tail-patching those entry points. 1990 was only about 2 years after quickdraw was re-written in C instead of assembler. Also, application developers made assumptions, e.g. sending F-line opcodes thinking any 68020 machine has an FPU (sorry!). So OO looked like the way out from that tangle.
Leaking tech docs were a big problem as Apple sought buy-in from partners. The "56" watermark might have overtly supported traceability back to the recipient. In ~1993 at Taligent we would also covertly vary variable names and such in sample code we delivered to different partners, after we found the code being shared anonymously.
Due to the OO scaffolding, the simplest application required implementing ~35 classes (yuck!), but the promise of modular intermixed code/edit/data (opendoc) was largely realized (yay!) before HTML and MIME types made complex data/display trivial (oh well).
As the length of the document shows, both Taligent and Copland were ... bedeviled with a million mid-level tyrants producing huge volumes of technical blabbage. Tremendous waste of brains, while a few sharp people were poking around Mach and finessing hardware abstraction layers.
Hoops (dev-env) and i18n seemed to be the only things that came out of that, and IBM pushed i18n into Java.