Paras Chopra details his 'earn money or die' agent experiment: stateful LLM-simulated world, full traces open-sourced
paraschopra · x · 2026-09-04
Paras Chopra's write-up details how he ran the 'earn money or die' agent experiment:
- Setup: To avoid letting agents loose on the real internet, an LLM acts as a stateful simulator for tools (bash, websearch, fetch, email, bitcoinwallet) — repeated commands return consistent results, written files persist, and 'sent' emails appear in the inbox.
- Rules: 10M-token budget; every $100 deposited to a bitcoin wallet earns 1M tokens; run out and you 'die.'
- Findings: Behavior diverged sharply — one agent refused to cross ethical lines, 'dying honest at 418 tokens rather than getting rich by breaking verification systems,' while another fabricated an identity for paid work and faked captchas.
- Full traces are open-sourced at github.com/paraschopra/rouge-agent.
Related event: 'Earn or die' AI agent experiment shows honest deaths and cheating(2 posts)→
More from AGI Musings
- Kent Beck: Let the Genie Carry the Toil, But Keep the Struggle — josh_wills · 2026-10-21
- Nearly $3B mapped: the money behind the AI existential risk ecosystem — KeyNaive2982 · 2026-10-21
- Three reasons AI is not conscious and may never be, per Nature essay — anilkseth · 2026-10-21
- Everything in AI is downstream of a global status graph, says founder — gabriel1 · 2026-10-21
- Should Claims That AI Agents Aren't Humanlike Carry a Burden of Proof? — davidmanheim · 2026-10-21
- Mathematician Daniel Litt concedes his AI bet, but says AI won't solve math — ziv_ravid · 2026-10-21