Three related hypothetical security flaws in Casper: does the protocol allow them?

[vbuterin]

@fresheneesz

The attacker wouldn’t slash himself on his own chain that he fully controls. So I don’t think you’re correct that the attacker loses money on his own attacker-chain when the attacker participates in the honest chain. Also, the attacker won’t lose money on the honest chain while his chain is kept secret. The only way I can see the attacker being slashed in the honest chain is if, when the attacker releases his chain, the honest group decides to rewrite history and slash the attacker’s accounts after-the-fact. This seems likely to raise the cost of attack to near 50% of the active stake, but is that currently part of the protocol?

There’s an additional security assumption that users reject chains that refuse to include slashing messages for more than 2 months. Non-censorship is a fundamental expectation that people have of blockchains, so it’s reasonable to make it a kind of validity rule.

See Censorship rejection through "suspicion scores" for more details.