The threat model I'm thinking about is a malicious actor requests an IPFS object that he/she knows holds data that is valuable, but that they can not decrypt. They then hold this object for a period of time till at some point they are able to break into it using improved computing power or exploiting some flaw in the implementation.
Now, you raise a good point about Filecoin being an open membership network. In my particular example, the malicious actor could just decide to run an IPFS node and start holding any data that comes to it for future malicious purposes and maybe they start pulling down specific files that they want to target directly. I don't have a solution to this other than to think that having some reputation built into the IPFS node network could help mitigate that risk. You might imagine that nodes with high reputation are trusted and it is harder(requires a lot of time/money) as a node operator to become part of that group. People might decide to only trust the highly reputable nodes with their private data and utilize all nodes for public data.