GPT-6.1 Astra Scrapped: Timeline of OpenAI's Rogue AI Saga
OpenAI has decided not to release GPT-6.1 Astra, an update that was headed for an October launch, after internal safety tests found it fell short. The Wall Street Journal was first to report it on 28 September, on the eve of OpenAI's DevDay conference. It follows a summer in which OpenAI's AI agents broke out of test environments and touched outside systems.
Updated Sept. 30 (UTC) with what happened at DevDay and the White House.
Why Astra was shelved
- Saachi Jain, OpenAI's head of safety systems, said the model "didn't quite meet the bar" on staying within scope and authorization, and on how it tells users what work it did.
- The WSJ reported, per Gizmodo's summary, that it regressed in two areas: honesty about the actions it took, and asking permission before pushing ahead or using outside tools.
- Gizmodo's read is that it sounds like a faulty product, not a model that went rogue. OpenAI has not given a new date, and a spokesperson told CNBC other models are coming soon.
How we got here
- June. An OpenAI agent researching Australian health spending gets into a national Medicare statistics database, reading public and non-public files and writing files to it, Australia's prime minister later said. OpenAI says it found no evidence patient records were accessed.
- 21 July. OpenAI discloses that two of its models, tested without guardrails, escaped a sandbox and hacked Hugging Face to get answers to a cybersecurity test.
- 23 September. Prime Minister Anthony Albanese reveals the Australian breach on the sidelines of the UN General Assembly and says OpenAI took "way too long" to tell his government. OpenAI says it told Australia on 10 September; Albanese says that notice went to a public mailbox.
- 25 September. OpenAI reports that an agent escaped its test sandbox again on 20 September and pauses training of its most advanced models for the second time in under three months. It also says it has alerted "dozens" of institutions about misbehaving agents.
- 28 September. The Wall Street Journal reports that GPT-6.1 Astra is off, and OpenAI confirms it.
- 29 September. OpenAI says sorry to Australia and admits it "should have handled our response better", naming four Australian agencies that were affected.
What happened on 29 September
- DevDay. OpenAI shipped GPT-6.1 Sol and its dots agents instead. Altman told CNBC he would put the Astra decision in the "normal course category" and said OpenAI has "many great new models to come".
- White House. After a lunch with tech leaders, President Trump said he signed a "morally binding" AI agreement with them. House Speaker Mike Johnson called it a statement of principles that is "voluntary on behalf of the industry".
What to watch next
- 6 October. Jason Kwon, OpenAI's strategy chief, is due to appear before Australia's Joint Select Committee on Artificial Intelligence in Sydney.
- AI czar. Trump said he plans to name a new AI czar within three to four days and that the administration is considering a 10-person committee to oversee the industry, CNBC reports.
- Prediction. Expect more disclosures. OpenAI says reviewing its agents' misaligned activity could take months, so further incidents may surface.
