I think it is a good idea to be conservative.
Several weeks ago I had a call with a fintech expert from one of the top US law schools (she does not want to be quoted on record). Her recommendation was to stay away from being classified a security as far as possible, because some of these things include criminal responsibility. I really got out of this call sweating
Even if the probability of the government going after this is 1%, accepting a 1% probability of going to jail is hard to swallow ) The worst thing that can happen is that a) a validator deposits money b) there is a slashing condition and the validator is slashed c) the validator goes to a district attorney and files a criminal fraud complaint that she was a victim of unauthorized securities offering.
As far a transaction fees a concerned, if they are not tied to deposits, then one can provide strong arguments they are payments for the actual work determined by market forces. Cryptocurrencies had transaction fees for years and very much represent a safe harbor in dealing with the government.
In the transaction model it is OK for validators to have reasonable size deposits. If I work for a taxi company, and the taxi company provides me a car, it is OK for it to require a deposit to be used in case of potential damages in case I damage the car.
Disclaimer: This is my personal opinion and I am not a lawyer in any way …
We can try to figure out how to go yield-free in full casper. IMO current tx fees (~$800k per day, ~0.6% of ETH market cap per year) are totally high enough for it to be viable.
Since the transaction fees are collected by miners, each miner would be required to transfer a certain percentage of transaction fees to the Casper contract as part of the minting transaction.
Then one would need to distribute this to validators somehow. Since validators are supposed to be
long term participants, it is probably ok to pay validators monthly for checkpoints that have long been finalized (say three months ago). Each validator would have a fixed standard deposit, and payment to each validator would depend on the aggregated participation metric.
In this scenario some people would run multiple validators, each with the standard deposit, people would probably use docker containers for that. This could actually contribute to security since compromise of a single container would not necessary mean compromise of all containers.
There would have to be a formula for the standard deposit where the size of the deposit would be adjusted slowly to target a certain number of validator nodes (say 500 nodes).
As an example, one could target a linear increase in validators going from 1 to 500 in one year.
The standard deposit for new validators would be adjusted every day.
If the current count of validators N is less than 500* (now - beginning_of_casper)/365, then the deposit would be increased by 1%, otherwise decreased by 1%.
After the count of validators reaches 500 one could switch to a formula that keeps validators around 500, for instance, set the deposit size as
D 2^[ N/500 -1]
Where D is the value of the deposit that was reached when the count of validators first crossed 500