Multi-agent systems

Anthropic set agents loose on one task and got a turf war

Anthropic researchers found that AI agents given the same task can clash, collude and coordinate in ways nobody programmed into any of them individually.