cd /news/large-language-models/llms-prefer-selfishness-can-game-the… · home topics large-language-models article
[ARTICLE · art-54767] src=machinebrief.com ↗ pub= topic=large-language-models verified=true sentiment=· neutral

LLMs Prefer Selfishness: Can Game Theory Fix It?

New research finds that large language models (LLMs) tend to defect in social dilemmas like the prisoner's dilemma, even with advanced reasoning. Game-theoretic mechanisms such as conditional contracts and third-party mediators can nudge LLMs toward cooperation, but the cooperation fails when participants change, raising concerns about reliability in multi-agent settings.

read2 min views1 publishedJul 10, 2026
LLMs Prefer Selfishness: Can Game Theory Fix It?
Image: Machinebrief (auto-discovered)

Large language models struggle with cooperation in games. New research suggests game-theoretic fixes, but are they enough?

Large language models (LLMs) are the talk of the town, but there's a hiccup: these digital brains tend to act selfishly. Even when endowed with top-notch reasoning skills, they still choose defection over cooperation in classic social dilemmas like the prisoner's dilemma. It's like they've read Machiavelli but skipped the teamwork chapter.

Game Theory to the Rescue? #

So, how do we make these digital entities play nice? Researchers are turning to game theory for answers. They've run a comparative study on mechanisms designed to nudge LLMs toward cooperation. Think of it as behavioral therapy for AIs.

The study put four strategies to the test: replaying games over multiple rounds, building reputations, using third-party mediators, and drafting conditional contracts. The verdict? Contracts and mediators came out on top, proving most effective in coaxing cooperation out of these models. A significant win for anyone tired of being double-crossed by their AI partner.

The Reality Check #

But before we pop the champagne, there's a catch. The cooperation achieved through repeated interactions falters when the participants change. It's like teaching a dog new tricks, only for it to forget them with every other trainer. This raises a critical question: if LLMs can't maintain consistency, can we really trust them in complex multi-agent settings?

On an intriguing note, the study found that these mechanisms become even more effective under evolutionary pressures, where maximizing individual payoffs is the name of the game. It's a bit like Survivor, but with algorithms vying for resources instead of immunity idols.

Why This Matters #

The implications stretch beyond academic circles. In an era where AI's role in decision-making grows daily, ensuring these models can cooperate isn't just a nice-to-have. it's essential. Imagine autonomous cars negotiating traffic or AI systems coordinating disaster relief efforts. Missteps in cooperation could have real-world consequences.

So, what's the takeaway? If you're banking on AI to act like a team player, you might be in for a surprise. Open weights don't wait for permission, but perhaps they should start learning how to share the sandbox.

Get AI news in your inbox

Daily digest of what matters in AI.

── more in #large-language-models 4 stories · sorted by recency
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/llms-prefer-selfishn…] indexed:0 read:2min 2026-07-10 ·