Developing language models that can deceive is a challenge for research on diplomacy, and there are potential benefits and downsides to the use of these AI agents in virtual worlds, including the issue of not knowing if you're playing against humans in online games.