Anubis looks very interesting - while I love some aspects of the AmigaOS user experience, it's apparent that some of the underlying system is woefully outdated and as far as I can tell none of the three existing systems have a clear roadmap for modernisation. MorphOS seemed to have the roadmap originally although as far as I can tell most development so far has been on the ABox. An OS that tries to place the best AmigaOS features on a modern foundation sounds like a great idea, however there's not a great deal of info on the Anubis website. So here are a few questions 
As far as I can tell from the website, Anubis is trying to build an AROS-derived userland on top of a Linux kernel - is that correct? I guess that's similar in many ways to what MS did with Windows NT, using the same Win32 API on a new kernel. OSX is another possible comparison, however as you mentioned the old-style Carbon API was only intended as a legacy solution.
Is there any particular reason why Linux has been chosen as the kernel? I always thought that AmigaOS and clones were much more microkernel based, but presumably the Anubis team considers a monolithic design a better choice for various reasons?
Basically it came down to which kernel and drivers were going to be the best option. One thing that has dogged AROS was the lack of drivers, especially gfx with hardware acceleration. Going with Linux means we get OEM drivers from nVidia and AMD/ATI to use with a huge number of devs working on new or updated drivers.
From the perspective of a user, how similar is the system going to be to AmigaOS/AROS/MOS? What bits of the interface/filesystem/etc do you think are worth keeping and what is likely to go?
For the end user, it's going to be as close as possible when it gets close to being full blown release. There is going to be a large gulf between where the initial release is to where the Devs want it at eventually though.
In terms of actual code, how much are you planning to re-use from AROS? From my highly limited knowledge of how Amigalike OSes work internally, doesn't much of the AmigaOS API rely on the fact that any task can quickly access the memory of another task, something that's not compatible with memory protection? Will this be a big issue?
Very little, if any, for the internal code. Running AROS via emulation is an option though. Basically the devs are taking the best from their learning experience with AROS and apply it to Anubis-OS. Again, AROS was more for Devs while Anubis-OS is aimed at the end user experience. The target of the end product is to hide as much of the ugliness of Linux from the end user as possible.
What about backwards-compatibility with existing AROS applications? Will AROS applications be able to access some of the new features, a bit like on OS4? Or is it intended for old applications to run in a sandbox, and build the next-generation stuff up separately (like the ABox/QBox idea in MOS)? Or would they just have to be ran in a VMed instance of AROS? Also, are you planning any integrated 68k emulation or would system-friendly 3.x apps need to be run in UAE?
Last I heard, but that was under the last management team so that may now change, UAE integration was going to happen.
Apologies for the insane number of questions, and I realise that the project's at a very early stage so it might not be possible to give detailed answers. However without knowing this kind of thing it's extremely hard to properly understand what you're aiming for.
It's under flux with new management so give the new manager a couple of weeks on how he wants his game plan to implemented. He is also attempting to bring in some new devs from the outside.
