Frontrunning bots have been exploiting the oracle service to read the next update and insert a transaction ahead of the update to create essentially risk-free profits.
The oracle is currently vulnerable to front-running. There are several measures in place to ensure that profitability from this is as low as possible, but more sophisticated bots have been written recently that have a much higher probability of profiting. In order to protect the system, we have implemented a mechanism where if the oracle detects it is being front-run it will front-run the front-runner with a tx to change the exchange fee rate to 99%. This is a credible threat to attempts to front-run the system to ensure even if they are profitable for a period there is a high likelihood they lose everything if the oracle detects this activity. There is the potential for collateral damage if a genuine trader executes a trade in the same block, but we are implementing protections against this so that a user can specify a fee rate above which the tx will be rejected. We expect to have this implemented soon.
In order for the system to remain viable, traders must not be able to frontrun the oracle. This feature ensures that there is a strong penalty for doing so that ought to discourage attempts to frontrun.
The oracle code is currently closed source, so we will not be publishing the exact spec of the mechanism. But above certain thresholds, if a bot executes a tx at a higher gwei than the oracle in the same block, the oracle will escalate to push an even higher gwei transaction to raise the exchange fee to 99% then restore the exchange fee rate back to its original setting.
UPDATE: We have modified the specification and are putting this SIP back into Proposed status, the Oracle will now call a non-public function setProtectionCircuit which will burn the entire balance of the trade rather than sending it to the fee pool. This mechanism is far more robust and targeted and should reduce potential collateral damage as it specifically allows the oracle to flag a single transaction and implement this function against only that transaction.
We have implemented several soft mechanisms to reduce the profitability of frontrunning over the last few months, and we plan to introduce more, but without a credible threat to profitability it is still the optimal strategy for bot developers to attempt to attack the system if the only penalty is reduced profits. If there is a risk of total loss of funds, this changes the strategic outlook for a potential bot developer significantly.
Copyright and related rights waived via CC0.