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 most capable models were paused. Its official wording is narrower: only training, evaluation, and inference *with tool-use* of those models remain paused.
Full reasoning
OpenAI’s Sep. 25, 2026 misalignment report uses materially narrower language than this post.
In the official report, OpenAI 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 being paused. The post removes the important qualifier “with tool-use (defined broadly)” and therefore overstates the scope of the pause.
OpenAI’s other public statements point the same way. On Sep. 6, 2026, OpenAI wrote that the earlier pause “did not halt all research: some workloads resumed under stronger controls, while others remained paused.” And on Aug. 18, 2026, OpenAI said it paused frontier-model inference in research clusters for runs that could execute code or use tools that could access the internet, while its largest planned frontier RL run remained on hold and smaller-scale work continued.
So the post’s wording is inaccurate because it turns a qualified, partial pause into an unqualified claim about all inference and training of OpenAI’s most capable models.
3 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.
- Research acceleration: The view inside OpenAI | OpenAI
This did not halt all research: some workloads resumed under stronger controls, while others remained paused.
- Pacing model development in an era of cyber-critical capabilities | OpenAI
Immediately following the OpenAI-Hugging Face incident, we paused frontier model inference in research clusters for runs that could execute code or use tools that could access the internet. Our largest planned frontier RL run remains on hold while we conduct smaller-scale training and evaluations.