Science & Tech · AI · 3 hrs ago
Researchers worry that sharing work with AI could expose their ideas
Scientists are using AI tools to review literature, refine writing and help develop mathematical proofs, but some worry about sharing unpublished work with AI companies.
Mathematician Tristan Buckmaster says an OpenAI model made progress on a problem after he had uploaded related work to Codex, and he found the timing suspicious.
OpenAI denies that his prompts influenced its system, though the company’s stated time window changed.
The problem involves the Navier-Stokes equations, which describe fluid flow and are linked to a $1 million prize.
Buckmaster still uses Codex but says he no longer puts ideas he wants to keep original into AI tools; researchers also worry that grant reviewers may upload unpublished proposals.
Unclear data practices are eroding trust, and researchers using personal accounts are advised to check settings and opt out of data use where possible.
Researchers are concerned that sharing unpublished work with AI tools could expose their ideas or allow companies to present similar results.
Mathematician Tristan Buckmaster says an OpenAI model reached a point in work he had shared with Codex shortly after the company learned his name.
OpenAI says Buckmaster’s prompts could not have influenced its system, including through training.
Researchers use AI for tasks ranging from literature reviews and editing to generating mathematical proofs.
Privacy expert Jennifer King says personal-account users may need to opt out of having their information used for training.
- Who
- Researchers, including mathematician Tristan Buckmaster, are concerned about sharing unpublished work with AI companies.
- What
- They fear that ideas shared with AI tools could be exposed or used in ways that lead companies to claim similar work.
- When
- The articles were published on October 12 and 13, 2026. Buckmaster says the model reached the point in question within a day of OpenAI learning his name.
- Where
- Buckmaster shared drafts through a personal paid account on OpenAI’s Codex. The articles also discuss researchers using AI tools more broadly.
- Why
- Researchers worry about data privacy and the lack of transparency around how AI companies use information shared with their systems.
Tristan Buckmaster
OpenAI
Connection between shared work and model progress
Tristan Buckmaster
Buckmaster says the model reached the point where he and his collaborator had stopped within a day of OpenAI learning his name.
OpenAI
OpenAI says it is impossible that Buckmaster’s Codex prompts influenced its system in any way, including through training.
Interpretation of timing
Tristan Buckmaster
Buckmaster calls the timing suspicious.
OpenAI
OpenAI rejects any connection between its system and his data.
Buckmaster and Levent Alpöge worked on a step toward solving the Navier-Stokes problem.
Buckmaster says the model reached the point where he and Alpöge had left off.
Buckmaster said he was not accusing anyone, according to the Yahoo Finanza article.
Yahoo Finanza published its report on Buckmaster’s concerns.
Scientific American published its report on researchers’ concerns about sharing work with AI.
- Researchers
- Use AI for literature reviews, editing and generating mathematical proofs
- Buckmaster’s tool
- OpenAI Codex
- Account type
- Personal paid account
- Research problem
- Navier-Stokes
- Prize mentioned
- $1 million
- OpenAI solution announcement
- September 8 (year not specified in the article)
What people are saying
Before they knew my name, they had made no progress. After they found out who I was, they suddenly picked up from where we had left off... It seems suspicious to me.
We can categorically state that it is impossible that Dr. Buckmaster’s Codex prompts over the last two months influenced our system in any way, including through training.
I’m not encouraging it, and I’m not doing it myself.
Right now, these systems infer things about you without telling you that they’re inferring things about you.
Sources
Related news
Anthropic warns AI models may soon spread across home devices
Developer releases free, open-source alternatives to Microsoft Office apps
Scientists call for stronger safeguards against AI hacking
Meta launches Muse Wonder AI models aimed at catching up with rivals
Dwarf Fortress creator Tarn Adams says he does not plan to use generative AI
Epic Games shifts toward AI to speed up game development
Microsoft plans large-scale rollout of Maia 300 AI chips from 2027
Unitree founder says AI advances could speed up robotics breakthrough
Samsara launches AI agent platform for physical operations
Nvidia tones down DLSS 5 for Hogwarts Legacy debut
Microsoft president calls for enforceable global AI safety rules
