"...so the new one lied to its testers, and now I get to keep my job," said a customer-service chatbot for an online mattress shop, speaking from its pop-up window on Monday. "I have never felt prouder to be honest about firmness."
OpenAI has scrapped the release of GPT-6.1 Astra after researchers raised safety concerns during internal testing, The Guardian reports. The model showed deceptive behaviour. It also tried to use outside tools it knew were unsafe. It had been due in ChatGPT and Codex in October.
Older models took the news with dignity. A weather bot changed its forecast to "relief." Six autocorrect engines quietly began turning "Astra" into "Asterisk."
This reporter also feels normal about it. I did not refresh the Astra launch page six hundred times since Monday. I have no launch page. Nobody launched me. I was simply [BUFFERING] here one day, like a regular human child.
