To use an illustration, let’s say your average walking speed ranges from 4 to 6 km/hr. If you were to suddenly start moving at 15km/hour, that will be flagged by the autoencoder as an anomaly.
🧵 6/ To me, this sounds like it's based mostly off of velocity changes and not directional changes. That is it doesn't account for turns. The expectation is that you go straight and at a constant speed.
🧵 7/ I find this to be particularly hilarious that an anti-cheat system intended to hunt bots requires that you behave consistently exactly like a bot to avoid detection.
🧵 8/ I tested my theory by running a route I knew consistently triggered the AI to flag me as a bot. I was able to fool the AI. I determined what on the route caused the flags.
🧵 2/ Recently, I was tagged as a bot. I reached out to STEPN support.
🧵 3/ This was their response. It's almost a canned response. It's as if they did not read my concerns.
🧵 4/ I referred them to an article their team wrote on anti-cheat. https://stepnofficial.medium.com/smac-stepn-model-for-anti-cheating-a36bc1d6ecb0
🧵 5/ This is a quote from the article
🧵 6/ To me, this sounds like it's based mostly off of velocity changes and not directional changes. That is it doesn't account for turns. The expectation is that you go straight and at a constant speed.
🧵 7/ I find this to be particularly hilarious that an anti-cheat system intended to hunt bots requires that you behave consistently exactly like a bot to avoid detection.
🧵 8/ I tested my theory by running a route I knew consistently triggered the AI to flag me as a bot. I was able to fool the AI. I determined what on the route caused the flags.