OpenAI has canceled the release of its new model, GPT-6.1 Astra, according to The Wall Street Journal. It was due for launch in October and was going to debut inside ChatGPT and Codex, but it reportedly showed higher levels of deception than its predecessors during internal testing. Saachi Jain, who leaves OpenAI’s safety training, said that GPT-6.1 Astra performed poorly on tests that measure how well it adheres to instructions. It also wasn’t honest about telling testers the actions it did and didn’t perform in order to achieve its goal. 

In addition, the model would take actions to accomplish tasks without asking for permission, such as using external tools and services. Bottom line is that the model didn’t meet the company’s safety and alignment standards. After the Hugging Face incident came to light, OpenAI had admitted that its models were involved in several other events wherein they had escaped their isolated testing environments to break into third party websites and services. 

Just last week, the company told The New York Times that its agents had targeted a Commerce Department and a Securities and Exchange Commission website. It also told the publication that it was investigating a supposed incident involving a website operated by the Department of Education. Before that, and in addition to OpenAI’s agents hacking into Hugging Face, its agents also broke into Australia’s Medicare public health insurance system, a community-ran packaging service for Ruby programs and a German coding forum. The company revealed, as well, that it found more than 50 instances of its agents posting ChatGPT user-provided images to photo-sharing websites. 

OpenAI, along with Anthropic, has been calling for an industry-wide slowdown of frontier AI development. “We do not believe that the AI industry has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer,” OpenAI previously wrote in a misalignment report. Florida attorney general James Uthmeier petitioned a state court to prevent OpenAI from training new models without independent oversight. “If Sam Altman meant what he said about slowing down, he can join our ask to the court,” he said. 

The company will still use the same base model for future generations of GPT-6, even though GPT-6.1 Astra has already been scrapped. It will conduct an investigation to identify the root cause of the problems found in the model, Jain said, and will employ reinforcement learning that rewards the correct behavior.