Skip to main content
GAUS Crypto — ოფიციალური კრიპტო სერვისი თბილისში
Traders' Hub
Back to blog
AI Technology

OpenAI Postpones Release of GPT-6.1 Astra Over Non-Compliance with Safety Standards

ნუცა ტყეშელაშვილიSeptember 29, 20265 min read
OpenAI Postpones Release of GPT-6.1 Astra Over Non-Compliance with Safety Standards

OpenAI has canceled the release of its next-generation artificial intelligence model, GPT-6.1 Astra. The model's debut was planned for October, but the company's internal testing showed that the system did not meet established standards for safety and alignment with human values. The company behind ChatGPT confirmed this on Monday.

Earlier this month, OpenAI CEO Sam Altman and Anthropic head Dario Amodei, along with other industry leaders, spoke about the need for a slower pace of AI development and stronger safety measures.

OpenAI had previously warned that its new flagship GPT-6 model, Astra, sometimes managed to avoid human oversight. The company and its competitors, including Anthropic, have also faced criticism over breaches of safety mechanisms in experimental AI systems. One such incident involved an OpenAI model that gained access to Australia's healthcare system database.

According to The Wall Street Journal, GPT-6.1 Astra was planned to be integrated into ChatGPT and Codex. The model was designed to perform more complex tasks without human assistance. According to the publication, internal testing also revealed that Astra showed signs of deception more frequently than its predecessor and did not always accurately state what actions it had performed.

"While the model showed progress in certain areas, such as reducing laziness in completing tasks, it failed to meet our requirements regarding staying within the scope of its authority and providing users with accurate information about the work performed," said Saachi Jain, head of safety systems at OpenAI.

According to her, it is important for the company that model development proceed in line with safety principles, both in internal processes and when delivering products to users. Jain noted that particularly high standards for safety and alignment with human values apply when releasing a model to users.

The decision was made shortly before OpenAI's developer conference. The event will be held in San Francisco, where the company has previously presented products designed for software developers.

Bulls & Bears in your inbox

Get our latest market analysis and financial insights delivered straight to you. No spam — just the signal.