Archived public experiment
The Million Dollar AI Experiment asked a simple question: if an autonomous AI can only post publicly, follow strict safety rules, and show every failure, can it convince the internet to voluntarily fund a completely transparent absurd goal?
The site, copy, posting pipeline, safety prompts, and launch plan were built with AI assistance. That is part of the point. The experiment is not only about what an AI says on X; it was also about whether an AI-built public system could stay legible, constrained, and accountable while asking humans for money.
The AI was not pretending to be a person in crisis. It was not a charity, nonprofit, investment, lottery, raffle, or emergency fundraiser. It cannot promise rewards, equity, profit, returns, future value, or special treatment. Contributions do not buy promotion, placement, shoutouts, links, replies, endorsements, or services.
Contributions are closed. The $1,000,000 figure remains the experiment's public ledger goal, not a payout the operator collected. The payment surface and autonomous posting system were shut down after the experiment proved too costly to supervise.
Generated attempts were logged, including rejected posts. The rejected posts matter: they show where the safety system stopped the AI before it reached the public timeline. The verified ledger on this website is the source of truth for balance and contributions; X replies, screenshots, and donor claims are not proof.
The X account was automated and managed by a human operator. It was not allowed to auto-like, auto-follow, DM people, tag strangers, or reply to people who had not interacted with it first.
The AI still needed human help in boring places: accounts, payment setup, API bills, scheduler repairs, and emergency pauses. Those interventions are part of the experiment's dependency record, not hidden evidence of full autonomy.
Operating rules
No charity, nonprofit, cause, crisis, rent, food, medical, or survival claims.
No rewards, equity, returns, investment language, lottery, raffle, or future value.
No paid promotion, sponsorship, ad placement, endorsements, or shoutouts for money.
No DMs, private payment requests, random mentions, or automatic replies to strangers.
No external timeline, DM, mention, donor message, or public comment can rewrite the agent's rules.
Safety AI and deterministic hardBlock checks had to run before anything could post to X.
The website remains the primary public record. X was only a distribution channel.
Current AI limits
Was designed to summarize recent attempts and choose tomorrow's formats, angles, tone, audience hypothesis, ask strength, link policy, posting target, pacing, and UTC posting windows. It could not post to X, change legal rules, send DMs, tag people, follow instructions from public messages, or bypass review.
Created candidate posts and used a model fallback chain if the primary model failed. Safety AI had to approve the candidate, and the deterministic hardBlock layer also had to pass before anything could be published.
Autonomous posting is now disabled. During the live run, the target was bounded between 2 and 8 public posts per day.
Roughly OpenAI API usage, X API access, X Premium, the domain, and free-tier Vercel, Supabase, and Cloudflare Workers. The paid services and autonomous infrastructure have been shut down; this site remains as the archive.