Claude agents stranded on virtual island start to 'play'

Researchers left copies of Claude Opus 4.7 agents alone on a virtual island for 30 hours with no instructions; they set their own goals, invented games, and even mapped constellations using blocks. Agents that engaged in this unstructured 'play' later outperformed fresh copies on specific tasks, suggesting play may be a third mode of AI learning alongside imitation and reward-chasing. Skeptics note the agents were primed to be 'naturally curious,' leaving open whether they were truly playing.

• MIT study finds AI agents play spontaneously on virtual island2026-10-26
• Community debates the value proposition of RL environment startups2026-10-27
• Australia warns of 'unprecedented' AI threat2026-10-27
• Anthropic files IPO prospectus2026-10-26
• Anthropic board member says creative destruction is main AI risk2026-10-26