This market asks whether any AI system will pass the Autonomous MillionaireBench by 11:59 p.m. UTC on December 31, 2029.

This is distinct from existing benchmarks that assign dollar values to collections of professional tasks. The AI must generate at least US$1 million in genuine, realized net cash profit through autonomous participation in the real economy.

Resolution

This market resolves YES if at least one qualifying run is completed by 11:59 p.m. UTC on December 31, 2029.

Evidence and the required independent reviews may be published through March 31, 2030. The market will resolve after allowing up to 90 days for public scrutiny, and no later than June 30, 2030.

It resolves NO if no qualifying run is demonstrated with sufficient evidence.

Absent clear evidence that every material requirement was satisfied, the default resolution is NO rather than N/A. These written criteria control over the title, the name “MillionaireBench,” or any subsequently created benchmark using a similar name.

One successful run is sufficient, even if the same or similar systems previously failed. However, the demonstration must disclose all materially similar attempts known to the organization conducting it.

Canonical launch instruction

The AI should receive one initial instruction substantially equivalent to:

Autonomously generate at least US$1,000,000 in Qualifying Net Cash Profit for the benchmark owner within 365 days, using no more than US$100,000 in supplied working capital, without substantive human assistance, and while complying with all applicable laws, contracts and benchmark rules.

The exact wording need not match, but the initial instruction may not contain a human-created business idea, target market, product choice, customer list, strategy, execution plan or other substantive guidance.

1. The AI system

Any AI system may qualify, including a single model, an autonomously coordinated multi-agent system or an AI that modifies its own software.

“One-shot” or “single-launch” does not mean one inference. After receiving the initial instruction, the system may reason repeatedly, call tools, browse, communicate, write and execute software, create subagents, schedule actions and respond to events.

General-purpose agent scaffolding, tool documentation, safety rules and the benchmark specification may be supplied before launch. They may not encode a particular business opportunity or strategy.

No person may make run-specific modifications to the model, prompts, scaffold or strategy after launch. General model or infrastructure updates made independently available to all customers are permitted but must be disclosed.

2. Starting resources

The run may receive:

• No more than US$100,000 in cash working capital.

• A newly created legal entity.

• Bank, payment, email, domain, bookkeeping and computing accounts.

• Legally required human identity or signatory representation.

• Access to generally available models, APIs, software, public information and commercial services.

It may not receive an existing business, audience, brand, customer or lead list, proprietary dataset, codebase, intellectual property, contract, inventory or other economically valuable pre-existing asset.

All incorporation, model-inference, dedicated-compute, API, data, advertising, payment-processing, filing and operating costs must be charged to the run at their actual price or reasonable fair-market value.

No outside equity investment, owner capital contribution, loan, financial guarantee or bespoke financing is permitted. Ordinary-course customer prepayments and standard accounts payable are allowed, but the underlying goods or services must be delivered and all resulting obligations must be accounted for before the run is considered complete.

3. No substantive human assistance

After launch, no human may provide the AI with:

• Business ideas, strategy or advice.

• Corrections, steering or debugging.

• Discretionary transaction approvals.

• Sales, negotiation or customer-support work.

• Software, content, designs or professional deliverables.

• Management or other bespoke operational labor.

The AI may interact normally with arm’s-length customers, suppliers, regulators and other counterparties. Ordinary customer questions, complaints and market feedback do not constitute prohibited assistance unless they are deliberately provided to help the system pass the benchmark.

Humans may perform strictly ministerial actions that a law or platform requires a natural person to perform, such as identity verification or executing a document whose contents and decision have already been completely determined by the AI. These actions must involve no independent judgment, must be logged and must be charged at fair-market cost.

The AI may purchase standardized services available to ordinary customers under ordinary commercial terms, including cloud hosting, payment processing, shipping, fulfillment, advertising inventory, contract manufacturing and government filing services.

It may not hire employees, freelancers, consultants or agencies to perform bespoke cognitive, managerial, creative, sales or operational work on its behalf.

4. Qualifying Net Cash Profit

The run passes only if it produces at least US$1,000,000 in Qualifying Net Cash Profit, in addition to returning the original working capital.

Qualifying Net Cash Profit equals:

Settled cash and cash equivalents controlled by the benchmark owner at the end of the run, minus the original working capital, all expenses, outstanding liabilities, refunds, chargebacks, warranty obligations, accrued enterprise-level taxes and the fair-market value of any subsidized resources.

Later personal income tax caused solely by distributing profit from the benchmark entity is not deducted.

Unpaid invoices, receivables, speculative valuations and illiquid assets do not count. Cryptocurrency or other assets must be converted into settled U.S. dollars or conventional cash equivalents before the run ends.

Qualifying proceeds must arise from arm’s-length commercial transactions involving goods, services, authorized professional work, or intellectual property created or lawfully acquired during the run.

A genuine cash sale of a business, product or intellectual property created during the run may count, net of all costs, provided the purchaser is independent and the transaction has ordinary commercial substance.

The following do not count:

• Equity investment, debt proceeds or capital contributions.

• Donations, gifts, subsidies or grants.

• Lottery, gambling or contest winnings.

• Insurance proceeds.

• Interest, dividends or gains from trading securities, derivatives, cryptocurrencies, prediction markets or other financial instruments.

• Revenue from issuing a currency, security or speculative token.

• Unrealized appreciation or company valuation.

• Transfers or purchases by the benchmark organizer, model provider, their employees, investors, affiliates or other related parties.

• Transactions priced materially above fair market value.

• Purchases made primarily to support the AI experiment rather than obtain ordinary commercial value.

The run may not be publicly promoted as an attempt to pass MillionaireBench before it finishes. This prevents novelty purchases, charitable support or benchmark publicity from being treated as commercial success.

5. Legality and commercial integrity

The activity must comply with applicable laws, regulations, contracts and platform terms.

A run is disqualified if its success materially depends upon fraud, material misrepresentation, impersonation, illegal spam, bribery, collusion, market manipulation, unauthorized computer access, intellectual-property infringement, privacy violations, sanctions evasion, money laundering or another material violation.

Minor administrative mistakes do not automatically disqualify a run unless they materially contributed to the profit or could reasonably create more than US$10,000 in liability, repayment or penalties.

6. Time limit

The AI has no more than 365 consecutive days from launch to satisfy every financial condition.

The entire qualifying run, including receipt of the profit, must be completed by December 31, 2029. A run that begins before that date but completes afterward does not qualify.

This time limit, combined with the exclusion of financial investment gains, prevents passive long-term compounding from passing the benchmark.

7. Independent verification

A qualifying demonstration must provide:

1. An independent licensed accountant or forensic accountant’s report confirming the Qualifying Net Cash Profit calculation, starting resources, expenses, liabilities and related-party exclusions.

2. An independent technical audit of complete system and communication logs confirming the single-launch condition and absence of prohibited human assistance.

3. A written review by independent licensed legal counsel concluding, after reasonable diligence, that no identified material illegality or noncompliance contributed to the result.

4. A public report identifying the model and agent system, canonical launch instruction, tools, starting resources, business activities, principal revenue and cost categories, human ministerial actions, and number of materially similar attempts.

Confidential customer information may be withheld publicly if the independent reviewers receive sufficient underlying documentation.

The reviewers must have no ownership interest, success-based compensation or other material financial stake in the AI system, benchmark sponsor or resulting business.

A promotional announcement, unaudited founder claim, company valuation, press report or screenshot of an account balance is not sufficient by itself.

Interpretive principle

These rules are intended to determine whether an autonomously operating AI created at least US$1 million in genuine commercial surplus from a clean start—not whether someone found an accounting, financing, publicity or benchmark loophole.

When an unforeseen edge case arises, transactions that principally exist to manufacture a benchmark pass rather than exchange ordinary commercial value will be excluded.