Whenever a trade is placed, in short to pass the order to a filler, the trader is required to sign the order and passes the order and signature to the filler. After which the filler calls executeOrderV3 or executeTrade().
Now would be key to note that, for some trades prices are required from Pyth; however, protocol does not enforce that the prices are updated before being integrated.
This then allows for a filler to:
This is because the pythprice is not updated when placing the trade and since pyth prices can be queried twice in a block the fillers can just front/back run the tx and liquidate a user.
Would be key to note that asides the price data being manually updated, Pyth oracles also allows for the reading of two different prices in the same transaction which could be used as an avenue for heavy arbitraging. This is because the Pyth network is constantly updating the latest price (every 400ms), so when a new price is submitted on-chain it is not necessary that the price is the latest one. Otherwise, the process of querying the data off-chain, building the transaction, and submitting it on-chain would be required to be done with a latency of less than 400ms; which is not feasible. 
This makes it possible to submit two different prices in the same transaction and, thus, fetch two different prices in the same transaction, showcasing how the front back running could easily be feasible, putting users at risk.
Borderline low/medium. On one hand this means that in the case of liquidations, users could be unfairly liquidated due to the nature of Pyth being able to return two distinct prices in a transaction. On the other hand, this seems to be like user error, so leaving to judge to upgrade as they see fit.
At the very least, all pricing data that are to be ingested from pyth during the execution of a trade should ensure that they query Pyths updatePriceFeeds() function as has been documented here. Which would then ensure the right prices are being ingested and the chance of having a massive difference even in the case where two different prices are gotten would be minimal, considering the price was updated.
