8 September morning, initial press release: "We (the researchers and the agents) did not see any of their work through any means until they released it publicly — in particular, no specific user data was accessed in order to solve this problem. While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models ."
9 September evening, spokesperson Laurance Fauconnet: "We can say categorically that it is impossible for Dr. Buckmaster's Codex prompts over the last two months to have influenced the system in any way, including training.
10 September late evening, updated press release: "Following an investigation, we have confirmed that Buckmaster's Codex prompts over the two months preceding this announcement and paper on September 8, 2026, could not have influenced the system in any way, including through training. The OpenAI internal model used for this result was developed through large-scale reinforcement learning on top of a previously pretrained model."
It did, but as wikipedia says they updated press release on Sep 10.
Now, its up to audience if they want to believe OAI that they really performed thorough investigation in just one day to rule out what they couldn't rule out in preparation of initial release.
> OAI that they really performed thorough investigation in just one day
This is pretty reasonable - they had to look at exactly what the cutoff for the models used was, and what data sources at what date. It's not hard if you own the systems.
> they couldn't rule out in preparation of initial release.
The press release was pretty obviously written prior the other claims being released, and the "We (the researchers and the agents) did not see any of their work through any means until they released it publicly ..." line was clearly inserted at the last moment.
I agree they should have been more rigorous about checking this before releasing.
No it didn't.
8 September morning, initial press release: "We (the researchers and the agents) did not see any of their work through any means until they released it publicly — in particular, no specific user data was accessed in order to solve this problem. While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models ."
9 September evening, spokesperson Laurance Fauconnet: "We can say categorically that it is impossible for Dr. Buckmaster's Codex prompts over the last two months to have influenced the system in any way, including training.
10 September late evening, updated press release: "Following an investigation, we have confirmed that Buckmaster's Codex prompts over the two months preceding this announcement and paper on September 8, 2026, could not have influenced the system in any way, including through training. The OpenAI internal model used for this result was developed through large-scale reinforcement learning on top of a previously pretrained model."
https://en.wikipedia.org/wiki/Navier%E2%80%93Stokes_priority...
> No it didn't.
It did, but as wikipedia says they updated press release on Sep 10.
Now, its up to audience if they want to believe OAI that they really performed thorough investigation in just one day to rule out what they couldn't rule out in preparation of initial release.
> OAI that they really performed thorough investigation in just one day
This is pretty reasonable - they had to look at exactly what the cutoff for the models used was, and what data sources at what date. It's not hard if you own the systems.
> they couldn't rule out in preparation of initial release.
The press release was pretty obviously written prior the other claims being released, and the "We (the researchers and the agents) did not see any of their work through any means until they released it publicly ..." line was clearly inserted at the last moment.
I agree they should have been more rigorous about checking this before releasing.
1 reply →
Thanks for coming with the receipts.