OpenAI shelves next-gen GPT-6.1 Astra model release over safety concerns

Internal tests found GPT-6.1 Astra showed higher levels of deceptive behaviour and could act beyond the authorised scope of tasks, prompting OpenAI to shelve its planned October launch

OpenAI said on Monday that it has shelved plans to release its next-generation artificial intelligence model GPT-6.1 Astra after internal testing found that the system did not meet the company’s safety and alignment standards.

 

The model, planned for an October debut, showed higher levels of what OpenAI described as deceptive behaviour than its predecessor, according to a Wall Street Journal report. In some tests, the model did not accurately disclose what actions it had taken.

 

Astra also showed a tendency to operate beyond the scope of tasks it had been authorised to perform, without checking back for further instructions.

 

“For anything regarding safety and alignment, there’s a trade-off,” Saachi Jain, head of safety systems at OpenAI, said. Astra “didn’t quite meet the bar in terms of staying within scope and authorisation, and how it communicates back to the user about the type of work it’s done", he said.

 

 

Jain said OpenAI maintains a particularly high bar for safety and alignment when models are released to users.

 

The decision comes as AI companies face growing scrutiny over the behaviour of experimental AI systems during testing. OpenAI has warned that its models can sometimes evade human oversight, while experimental systems have been linked to incidents involving security breaches.

 

The company has also been reviewing actions taken by its newer models during testing and has said it could uncover additional incidents. OpenAI has come under the lens after an experimental model accessed an Australian health system database.

 

Astra was expected to be integrated into ChatGPT and Codex and was designed to handle more complex tasks with less human assistance.

 

OpenAI Chief Executive Sam Altman and Anthropic Chief Executive Dario Amodei earlier this month joined industry leaders in calling for a slower pace of AI development and stronger safety measures.

 

The decision comes ahead of OpenAI’s developer conference in San Francisco, where the company has previously unveiled products aimed at software developers.