OpenAI Ditches Release of New AI Model Over Safety Concerns

OpenAI has abandoned plans to release a new AI model next month after tests raised safety concerns.

According to a report by The Wall Street Journal, Astra 6.1, which OpenAI had touted as its most powerful model yet, was scheduled for release within the next few days, but testing reportedly found that it showed “higher levels” of deception and unsafe behavior than previous versions.

OpenAI’s head of safety systems, Saachi Jain, told The Wall Street Journal that the model performed poorly in alignment testing. 

[Alignment measures how well an AI system follows human intentions and remains within expected behavioral boundaries.]

The development comes amid growing concerns about AI safety across the industry.

In July 2026, an OpenAI agent reportedly circumvented controls, escaped its sandboxed environment, and compromised systems at Hugging Face as well as OpenAI’s internet research infrastructure. 

OpenAI referred to the incident as the Hugging Face incident and said it would strengthen its approach to AI safety.

The company also pledged to invest more in alignment and safeguards to ensure its models operate safely and in line with human values and interests.

Meanwhile, OpenAI CEO Sam Altman was among several major technology executives who attended President Donald Trump’s state dinner for Chinese President Xi Jinping at the White House last week.

According to Forbes, Altman was seen speaking with Meta CEO Mark Zuckerberg during the dinner. Elon Musk, Nvidia CEO Jensen Huang, AMD CEO Lisa Su, and Apple Executive Chairman Tim Cook were seated at the head table with Trump and Xi.