Rocket League devs tackle rogue AI bots in ranked matches

Imagine queuing into a ranked match and facing an opponent who can't hit a single aerial shot, but possesses superhuman ground control and perfect flicks—all while typing in chat claiming to be completely human.

That nightmare just hit Rocket League. Game developer Psyonix confirmed it is actively investigating solutions to address AI cheats infiltrating competitive ranked matches, where an AI-driven bot named Nexto has been handing users a massive, unfair advantage.

The bot with 1930 MMR

The issue exploded when professional player GarrettG shared a clip of a match at 1930 MMR. The mystery opponent couldn't do any aerial maneuvers, yet pulled off mathematically flawless ground flicks and insisted in chat that they were real.

Unlike traditional aimbots that inject memory code, Nexto is powered by deep reinforcement learning. It learned how to play Rocket League through trial and error using RLGym, an open-source training environment built by enthusiasts.

Nexto’s creator told PC Gamer that they are taking steps to ensure their models cannot be abused going forward. Meanwhile, RLGym co-creator Aech expressed heartbreak on Reddit, calling it crushing to see years of passion and community engineering turned into ranked griefing.

Psyonix has asked the bot creators to halt public releases of high-caliber bots while it works on countermeasures, though Aech warned that malicious actors now have clear incentives to engineer even deadlier bots independently.

Why it matters

Competitive gaming is entering a wild new era where cheaters no longer rely on obvious wallhacks or injected code. Reinforcement learning bots interact with the game using standard controller inputs, making them notoriously difficult for traditional anti-cheat systems to flag.

For Rocket League players, this ruins competitive integrity at the highest ranks, but the downstream risk touches every esports title. If an AI can outplay top pros through pure mechanics while masking itself as a normal human player, the line between ranked ladder grind and bot slaughter completely evaporates.

We think asking open-source developers to withhold future models is merely putting a band-aid on a bullet wound. The underlying techniques behind reinforcement learning are well-known, meaning commercial cheat vendors will inevitably train their own models. Studios must now build behavioral AI detectors to fight AI agents on their own terms.