This week’s headlines called it a “delay,” a “scrap,” a “pause” and a “stop” — but OpenAI made four separate moves between Sept. 25 and Sept. 29, each with its own source and date. Here is what each one actually is.
- The pause (Sept. 25): OpenAI paused training of its most capable models.
- The canceled release (announced Sept. 28): OpenAI will not release a model called GPT-6.1 Astra.
- The rules proposal (Sept. 28): OpenAI published proposed “safety cases” that would be required before frontier training runs.
- The launch (Sept. 29): OpenAI released a different model, GPT-6.1 Sol.
The pause
OpenAI’s Sept. 25 report, cited in Plainly Now’s Sept. 27 report on the pause, said “all training, evaluation, and inference with tool-use (defined broadly) of our most capable models remain paused.” The company said it would resume “only when we are confident that we have additional safeguards” and expects to “hit pause” again in the future. Nothing in the OpenAI pages checked for this article says training has resumed, and OpenAI has not said when it will.
The canceled release
OpenAI said Monday it will not release GPT-6.1 Astra, a decision first reported by The Wall Street Journal, according to Al Jazeera. In a statement provided to Al Jazeera, Saachi Jain, OpenAI’s head of safety systems, said the model failed company standards for “scope and authorization, and how it communicates back to the user about the type of work it’s done.” She said, “For anything regarding safety and alignment, there’s a trade off.” Separately she said: “But when we ship it to users, we have an extremely high bar in terms of safety and alignment.” No OpenAI web page names GPT-6.1 Astra.
The rules proposal
In a Sept. 28 post titled “Towards safety cases for frontier AI training,” OpenAI proposed that structured “safety cases” should be required before continuing any “frontier reinforcement learning training run.” Among the elements OpenAI says a safety case should or could include: senior leaders who “should each have the ability to veto the run in order to ensure there are multiple internal checks on the run (e.g., research org lead / VP, Head of Safety, and Chief Scientist)”; set procedures and deadlines for pausing all covered runs; automatic pausing of runs when alerts go unacknowledged overnight; immutable transcripts; safety features designed to “fail closed”; a misalignment on-call role that can page executives; a written dissent process from another team; and public disclosure of investigation results once an investigation concludes.
The post calls safety cases “an aspirational north star we are building towards” and says the operational guidelines described “are in the process of being implemented at OpenAI,” adding: “We expect our practices to continue to evolve over the coming weeks.” The post says it focuses on frontier training and that deployment requires considering “a much broader set of alignment properties.” Many elements are listed as things a safety case “could include.” The post does not mention Florida’s lawsuit or the GPT-6.1 Astra decision.
The launch
On Tuesday, OpenAI released GPT-6.1 Sol, an upgrade to GPT-6 Sol. It is available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users, and through the API as gpt-6.1-sol, priced at $2 per million input tokens and $10 per million output tokens. A system card addendum, OpenAI’s safety document for the model, said that under its Preparedness Framework, OpenAI’s system for assessing model capabilities and safeguards, GPT-6.1 Sol is treated as “Critical in cybersecurity and High for Biological and Chemical capability,” and that it “uses the same safeguards stack as GPT-6 Astra.” GPT-6 Astra is an earlier OpenAI model; the canceled one is GPT-6.1 Astra.
Outside pressure
Florida Attorney General James Uthmeier filed a motion Sept. 28 for a temporary injunction against OpenAI entities and Sam Altman, in a lawsuit Florida filed against OpenAI and Altman in June, according to PYMNTS. The motion asks the court to bar OpenAI from “developing new artificial intelligence models without third-party approved safety guardrails” and from having ChatGPT “solicit engagement.” It argues: “They have asked the government to tie them to the mast.” An OpenAI spokesperson, in an emailed statement to PYMNTS, said the company had paused training of its most capable models and that “Governments have an important role to play in setting robust safety standards for AI, and we’re committed to working with Florida and other states on advancing pragmatic AI policies that apply to the entire AI industry — not just one company.”
Also Tuesday, Altman said in a CNBC interview, according to the AP: “We are pacing our progress, which includes sometimes not training a model,” and “We have to do these in a way where we can make very confident safety cases and claims, so that people don’t have a bunch of anxiety.”
What’s next
Florida’s injunction motion is pending before the Circuit Court of the Tenth Judicial Circuit. OpenAI has not said when the paused training will resume, and the safety-cases proposal remains a framework the company says is still being built out.