If OpenAI or Anthropic announces it has achieved recursive self-improvement before Jan 1, 2027, then the market resolves to Yes.
The statement must describe an achieved capability in which the company's AI autonomously conducts the research and development needed to materially improve the general capabilities of its most capable or near-frontier AI systems, or produce a more capable successor, with the improved system able to drive further such improvements.
Toy, narrow, sandboxed, proof-of-concept, prompt, skill, or benchmark-only improvements do not qualify. Humans may provide high-level goals, resources, oversight, and approval, but must not supply the substantive research choices or technical solution.
The capability must be affirmatively described as currently achieved in an official company communication or by an executive or research leader speaking about the company's capabilities. Future plans, speculation, offhand remarks, interviewer characterizations, leaks, model outputs, and third-party claims do not qualify.
A public release, named model, technical demonstration, or independent verification is not required.
- opens
- Sep 17 · 4:00 AM UTC
- closes
- Dec 31 · 6:59 PM UTC · may close early
- expires
- Jan 8, 2029 · 5:00 AM UTC
- category
- AI
- event
- independent outcomes