Experts Warn Against Sharing Math Research With AI After OpenAI Data Dump

The story so far

2 publishers have covered this in 2 articles since 12 Oct 2026. Quartz published first, on 12 October at 12:54 UTC; Latest from Tom's Hardware followed 23h 35m later. The most recent article, from Latest from Tom's Hardware, appeared on 13 October at 12:30 UTC.

Outlets reporting it: Quartz, Latest from Tom's Hardware.

3 prediction markets price this story:

• Will the highest score achieved by an OpenAI model on Humanity’s Last Exam in 2026 be 60% or higher? — the market puts it at 38%, with $3k traded in the last 24 hours, closing 1 Jan 2027. The market resolves to Yes if an OpenAI model scores 60% or higher on Humanity’s Last Exam by December 31, 2026, at 11:59 PM ET. The score must appear as “HLE Accuracy” or a clear equivalent on the official leaderboard at agi.safe.ai. Readers often mistake other accuracy metrics or news reports for the final resolution data.
• OpenAI — the market puts it at 8%, with $3k traded in the last 24 hours, closing 31 Dec 2026. The market resolves to Yes if OpenAI holds the number one spot on LiveBench.ai's Coding Average ranking on December 31, 2026. Otherwise, it resolves to No. Readers often mistake the date of the linked news story for the resolution deadline.
• Will OpenAI have the best AI model at the end of October 2026? — the market puts it at 1%, with $3k traded in the last 24 hours, closing 31 Oct 2026. The market resolves by checking the arena.ai Text Arena leaderboard on October 31, 2026, at 12:00 PM ET. OpenAI wins if its model holds the highest rank among eligible models at that specific time and source. Readers often mistake the check date for the end of the month or overlook that auto-evaluated models are excluded from the ranking.

Odds are the market's probability that the outcome happens, updated every few minutes. They are information about what traders expect, not a forecast we make and not investment advice.