AI that can read compiled code and reconstruct it is drawing two opposite reactions in the games world. On one side, according to PC Gamer and Kotaku, Digital Sun Games, the studio behind Moonlighter, cancelled a public Steam playtest of its next game, ReVamp, for fear that models like Claude Opus 5.5 could decompile the game's code and build a copy from it. "We know we're not the only developers feeling this way," the studio wrote.
On the other side: refusals
At the same time, according to reports, users who rely on Claude for reverse engineering, taking software apart to understand how it works, have complained since October 21 that the model starts a task and stops midway because of its safety system. Those affected include developers trying to fix or modify games and smart TVs, and according to some, entirely legitimate tasks too.
What Anthropic says
Anthropic has not announced a policy change. According to one report, the company confirmed that the model has cybersecurity filters but did not announce a specific ban on reverse engineering. The picture is not clear-cut either: some users report that everything works normally.
The dilemma
Reverse engineering is a basic tool for security researchers, who take malware apart to understand it, and for developers preserving old software. But the same capability is used to clone products and find exploitable flaws. The stronger the models get, the thinner the line between legitimate help and enabling harm, and AI companies struggle to draw it.
