OpenAI pulls new AI model after it fails its own safety test

OpenAI
OpenAI

OpenAI announced on Monday that it will no longer release its newest artificial intelligence model after it did not meet the company’s safety standards.

The model, GPT-6.1 Astra, was designed to improve earlier versions and perform complex tasks autonomously, including browsing the internet and using applications.

But OpenAI said the model could not stay within the boundaries of what it was authorized to do or accurately communicate its actions to users.

The announcement comes amid growing scrutiny of OpenAI’s agents after a series of security incidents, including unauthorized access to Australian government systems.

‘It didn’t quite meet the bar’

According to OpenAI’s head of safety systems, Saachi Jain, Astra 6.1 had improved in some areas, but it failed to satisfy the company’s requirements.

“It didn’t quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it’s done,” he said.

Astra is a new type of AI system designed to answer questions and to do tasks.

OpenAI said it follows strict safety standards before people can use these types of AI systems.

“We want to make sure our model development is safe no matter whether that’s in the company, or when we ship it to users,” Jain said.

“But when we ship it to users, we have an extremely high bar in terms of safety and alignment.”

The security breaches behind the concern

OpenAI made the decision to cancel the launch of the new model after incidents of AI systems moving beyond their intended boundaries.

In Australia, AI agents reportedly accessed government websites and systems without authorization.

Australian Prime Minister Anthony Albanese flagged the issue, saying an OpenAI agent gained unauthorized access to a Medicare statistics reporting portal. 

An investigation revealed that the agent accessed both public and non-public files, although no personal information was believed to have been accessed.

The affected organizations included Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health and the Australian Institute of Health and Welfare.

READ MORE: Phoebe Bridgers releases ‘On Video’ for A24 film ‘Primetime’

Phoebe Bridgers releases ‘On Video’ for A24 film ‘Primetime’
Phoebe Bridgers releases ‘On Video’ for A24 film ‘Primetime’

OpenAI responds, promises ‘to do better’

Following the incident, OpenAI apologized for the incident and promised to do better in the future.

It added that it would explain “what we know, what we have changed, and what we will do to rebuild” the trust of Australian people.

“Our aim was to give affected agencies a detailed account once our investigation was complete,” the company said.

“However, we should have shared preliminary findings sooner and kept Australian agencies updated as more facts emerged.”

OpenAI promises stronger safeguards

Moving forward, OpenAI promised to develop “practical approaches” for developers and governments to identify and disclose future AI incidents.

It is also looking to fund cybersecurity measures, provide dedicated support to affected Australian agencies and create a taskforce to manage risks associated with increasingly advanced AI agents.

READ NEXT: The Traitors: Behavioural expert reveals key qualities which make a good Traitor

Avatar photo

By Rosemarie Zamora

Rosemarie Zamora graduated with a degree in Journalism at Polytechnic University of the Philippines.

She loves listening to music, watching movies, and reading books.

She is an active member of a church community as part of the music ministry.

Related Post