When initializing a dictionary with dict.fromkeys, if an object is used as initial value, then the value is taken by reference instead of by value.
This results in incorrect behavior since Python doesn't have "by reference" or pointers by default.
Attached file shows an example and one work-around.

This is the same basic problem seen with sequence multiplication, mutable default arguments to a function, etc. It's not going to change though; the documentation says it "Create a new dictionary with keys from iterable and values set to value." "set to value" doesn't allow for implicit copy operations (shallow or deep), and trying to support it would slow fromkeys and violate the normal expectations for functions that reuse an input argument. Essentially, the problem you have is one of expectations; the behavior is correct, but perhaps the documentation could be clearer.

In any event, the general solution to your problem is a dict comprehension (or for dict subclasses, passing a generator expression to the subclass constructor), replacing the incorrect:

    dict.fromkeys(iterable, [])

with the similarly concise/straightforward:

    {k: [] for k in iterable}

or (for general use, where dict can be changed to a dict subclass):

    dict((k, []) for k in iterable)

I've marked this as a documentation bug if someone wants to take a stab at making the behavior more clear.

New changeset da63b321f63b697f75e7ab2f88f55d907f56c187 by Raymond Hettinger in branch 'master':
bpo-36715: Add usage note for dict.fromkeys() (GH-12974)
https://github.com/python/cpython/commit/da63b321f63b697f75e7ab2f88f55d907f56c187