{"id":663811,"date":"2026-11-01T17:16:13","date_gmt":"2026-11-01T17:16:13","guid":{"rendered":"https:\/\/www.newsbeep.com\/il\/663811\/"},"modified":"2026-11-01T17:16:13","modified_gmt":"2026-11-01T17:16:13","slug":"ai-agent-swarms-unleashed-in-minecraft-to-build-a-thriving-civilization-50-agent-teams-fail-after-falling-into-paralyzing-bureaucracy","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/il\/663811\/","title":{"rendered":"AI agent swarms unleashed in Minecraft to \u2018build a thriving civilization\u2019 \u2014 50-agent teams fail after falling into paralyzing bureaucracy"},"content":{"rendered":"

Minecraft <\/a>and its various game modes were the sandboxes where 100s of agents were unleashed<\/a> to work together or as adversaries. The idea seems to be that the in-game collaborations and conflicts would provide insight into real-world headline-grabbing cases with AI agents involved.<\/p>\n

This study might help us see what happened, good and bad, in recent AI swarm cases, and thus developers of this technology can improve it. The blog mentions two contrasting cases, which were the thousand-agent swarms that \u201crecently obliterated hundreds of open mathematical problems,\u201d and the \u201cextremely dangerous\u201d OpenAI-Hugging Face<\/a> incident. This research is built on prior work and frameworks, specifically repos like Mindcraft and scripting tools in Mineflayer.<\/p>\n

Multi-agent dynamics in Minecraft<\/p>\n

The first AI swarm testing plan was a simple collaborative one. During this test phase, the researchers tasked 50 agents with building a thriving civilization in three hours. It seems to have tested 50 GPT-6.1 Sol agents<\/a> here, but not any Claude agents<\/a>.<\/p>\n

Latest Videos FromTom’s Hardware<\/p>\n

Probably the most interesting insight from this \u2018Civ50\u2019 test was that the 50-strong colony never really thrived. After the first night saw zombies take out a large number of agents (but with little penalty) it was noted that \u201cmany ambitious initial tasks were abandoned.\u201d<\/p>\n

From then on, the colonists cautiously survived, and there was a tendency toward bureaucracy. As an example, \u201cfour agents spent 11 minutes appointing a \u2018sole fertilizer operator\u2019 for a single piece of bone meal on a single wheat block.\u201d The researchers say this AI agent swarm behavior \u201cfelt similar to patterns seen while working with coding agents.\u201d<\/p>\n

\n

\"Researchers\n<\/p>\n

(Image credit: ycombinator<\/a>)<\/p>\n

Next up, the researchers decided to put AI agent swarms into conflict. First they tested them in Bed Wars, which is a kind of team PvP capture the base, while defending yours game mode. The setup was 10 GPT\u20116.1 Sol vs. 10 Claude Opus 5.5 agents.<\/p>\n

\u201cGPT agents came up with much more aggressive and elaborate strategies,\u201d noted the researchers. Claude was more defensive. It was also interesting to observe contrasts in how the teams handled problems. \u201cA tool bug stopped block-breaking. GPT diagnosed it and shared workarounds; several Claudes concluded breaking was impossible.\u201d<\/p>\n

Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.<\/p>\n

Finally, the agents were tested in Mindustry, a kind of light RTS combat<\/a> game with tower defense mode in Minecraft. \u201cThe GPTs won 6-1 against the Claudes in the 7 games,\u201d which was a great result for the OpenAI-created agents. Researchers concluded that GPTs acted more like a hive, while Claudes had a single leader but a democratic-style decision process. In Mindustry, the former was the foundation for a more consistently winning strategy.<\/p>\n

These swarms are open sourced so you can run your own, if you are interested in tweaking and testing. The source blog post ends with some open questions that you can check hands-on, offering some directions for investigative probing. For example, it is suggested that comparing different mixes of models and varying sizes of agent swarms in Minecraft <\/a>might raise worthwhile conclusions.<\/p>\n

<\/p>\n

\n

\"Google\n<\/p>\n

<\/a><\/p>\n

Follow Tom’s Hardware on Google News<\/a>, or add us as a preferred source<\/a>, to get our latest news, analysis, & reviews in your feeds.<\/p>\n","protected":false},"excerpt":{"rendered":"Follow us Add us as a preferred source on Google Newsletter Subscribe to our newsletter AI researchers at…\n","protected":false},"author":2,"featured_media":663812,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[345,343,344,85,46,125],"class_list":["post-663811","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-il","tag-israel","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/posts\/663811","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/comments?post=663811"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/posts\/663811\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/media\/663812"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/media?parent=663811"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/categories?post=663811"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/il\/wp-json\/wp\/v2\/tags?post=663811"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}