is it possible to avoid matching literals in a string and only match at the start?

Instead of a complex regex for split() to separate the tokens you want, it’s often easier and simpler to just capture the tokens you want:

String input = "[02 Nov 2020 17:31:00,117] [12' for queue: 'weblogic.kernel.Default (self-tuning)'] [            ] [0EyKACgWsiME2VoobHq2EuBc49xq_NsuBmnV0dhrgPWrKotJgQ8-!-438943902!1604338067478] [Server] DEBUG tm.BitronixTransactionManager - shutting down journal - resetting servers";

for (Matcher m = Pattern.compile("\\[[^\\]]*\\]|-.*|\\S+").matcher(input); m.find(); ) {
    System.out.println(m.group());
}

Output

[02 Nov 2020 17:31:00,117]
[12' for queue: 'weblogic.kernel.Default (self-tuning)']
[            ]
[0EyKACgWsiME2VoobHq2EuBc49xq_NsuBmnV0dhrgPWrKotJgQ8-!-438943902!1604338067478]
[Server]
DEBUG
tm.BitronixTransactionManager
- shutting down journal - resetting servers

Explanation

\[[^\]]*\]    Match a `[ ]` enclosed token
|             OR
-.*           Match a `-` and the rest of the input (excl. linebreaks)
|             OR
\S+           Match a block of non-space characters (must be last)

The \S+ must be last, otherwise it will consume the [ or - characters that are lead-in to the other two token types.

This should also perform much better, because there is no lookaheads or lookbehinds, and there is little to no backtracking.