OpenAI vs Mathematician: What Does This Tell Us?
OpenAI vs Mathematician: What Does This Tell Us?
On October 6, 2026, OpenAI released a new collection of mathematical results produced by an internal AI model. The consequences for scientific research could be enormous. OpenAI’s announcement
Whether this transformation is good or bad for mathematicians is not the question we want to explore here.
There is another question: what happens when the company you trust with your unfinished work also becomes your competitor?
The dispute between OpenAI and mathematician Tristan Buckmaster makes that question difficult to ignore.
When your AI provider becomes your competitor
Buckmaster and his collaborator, Levent Alpöge, had been working on mathematical problems involving fluid motion. They used AI tools throughout their research, including Codex.
In his public account, Buckmaster explained that their project drafts had passed through Codex. When OpenAI described a related breakthrough, the research direction and timing concerned him. He asked whether their private sessions had influenced OpenAI’s system.
He also described a troubling exchange. After saying he would go public about their discussions, Buckmaster says OpenAI researcher Sébastien Bubeck responded:
“Why would you ruin your career?”
That is Buckmaster’s account of the conversation. Read his statement
Bubeck subsequently acknowledged making a career-related remark and apologized for his wording. He disputed the threatening interpretation, saying he intended to express concern rather than apply pressure. Read Bubeck’s response
A discussion about scientific credit should not become a discussion about whether speaking publicly might damage someone’s career.
OpenAI investigates OpenAI
OpenAI says its investigation established that Buckmaster’s Codex prompts from the preceding two months could not have influenced the system, including through training. OpenAI’s response
But this is precisely where the trust problem appears.
OpenAI operates the infrastructure. OpenAI controls the relevant internal records. OpenAI conducts the investigation and publishes its conclusion.
The researcher does not have the same visibility.
This is the judge-and-jury problem. A company’s explanation of its own conduct is not the same thing as independent verification.
The uncomfortable arrangement remains: a researcher can pay a company to help develop ideas while that same company pursues discoveries in the same field.
That deserves scrutiny, not just reassurance.
“Not used for training” does not mean “stays on your computer”
Training controls matter. But they answer a different question from where your work is processed.
OpenAI offers training opt-outs for individual services and excludes business data from training by default. Those protections do not turn a cloud service into a local model. OpenAI’s data-use policy
A desktop interface does not change that distinction. Codex can execute tools locally while sending relevant content to OpenAI for inference. Claude Cowork also uses cloud processing. OpenAI’s documentation, Anthropic’s documentation
ChatGPT, Codex and Claude Cowork can be extraordinarily useful. But using their cloud-based workflows still means entrusting another organisation with the material you submit.
For unpublished research, confidential documents or the beginnings of a business, that is not a minor detail.
Powerful AI should not require surrendering confidentiality
We should not have to choose between useful intelligence and keeping our work private.
The future we want is one where powerful models can help us think, research and create without requiring our unfinished ideas to leave our devices.
Not just a promise about what happens after an upload. The ability to avoid that upload entirely.
At Fenn, we apply that principle to working with your own files. Your index stays on your Mac or your chosen storage drive, and the default AI models run locally. Once the required models are downloaded, local search and AI processing can work offline. Optional cloud providers and external AI integrations are separate choices with different privacy implications. How Fenn handles privacy
This story is not a reason to reject AI.
It is a reason to demand more control over how we use it.
Your ideas should not need to leave your computer just because you want help developing them.