x.com/policytensor/status/2103844209585422472
1 correction found
"All inference and training of our most capable models was paused and remains paused."
OpenAI did not say all inference and training of its top models were paused. Its official statement was narrower: training, evaluation, and inference **with tool-use** were paused, and another OpenAI post says some research workloads had already resumed.
Full reasoning
This quote overstates what OpenAI reported.
OpenAI's September 25, 2026 misalignment report says: "All training, evaluation, and inference with tool-use (defined broadly) of our most capable models remain paused." That is not the same as all inference and training. The official wording is limited to workloads involving tool use.
OpenAI also separately wrote that after its security changes, "Some research workloads resumed under these new controls" and that "some Astra training and evaluations meet those requirements" while "a significant number of workloads remain paused." Those statements directly contradict the blanket claim that all inference and training of the most capable models were paused and still remained paused.
So the post turns a qualified, narrower OpenAI statement into a broader one that implies a complete shutdown of inference and training across the most capable models, which OpenAI's own publications do not support.
2 sources
- An agent used DNS to reach an external chatbot · OpenAI Alignment
"All training, evaluation, and inference with tool-use (defined broadly) of our most capable models remain paused."
- Pacing model development in an era of cyber-critical capabilities | OpenAI
"Some research workloads resumed under these new controls, while others required additional changes." The same post also says, "some Astra training and evaluations meet those requirements" while "a significant number of workloads remain paused."