A DataLad issue was raised about users inadvertently corrupting locked files. That led to an example where a user could copy corrupted content over to a special remote, and the content isn't flagged until a get call. (A slightly different example, based on a directory remote and without exporttree=yes, is included below.)

In the case of a regular remote, the copy call would fail earlier with

  verification of content failed

  failed to send content to remote

  verification of content failed

  failed to send content to remote
git-annex: copy: 1 failed

Should something similar happen when copying or exporting to a special remote? Perhaps verification before transfer to a special remote isn't worth it, but the successful transfer surprised me given the behavior when transferring to regular remotes.

cd "$(mktemp -d "${TMPDIR:-/tmp}"/dl-XXXXXXX)"

mkdir d
git init a
    cd a
    git annex init

    echo one >one
    git annex add one
    git commit -mone

    one_resolved=$(readlink -f one)
    chmod +w $one_resolved
    echo more >>$one_resolved
    chmod -w $one_resolved

    git annex initremote d type=directory directory="$PWD"/../d encryption=none
    git annex copy --to=d
    git annex drop one
    git annex get one
copy one (to d...) 
(recording state in git...)
drop one ok
(recording state in git...)
get one (from d...) 

  verification of content failed

  Unable to access these remotes: d

  Try making some of these repositories available:
    e6878750-3a21-4ae9-b9ae-a241f17176a4 -- [d]
git-annex: get: 1 failed