ETNET NEWS AGENCY (Nov 29) - OpenAI has canceled its planned October release of its new artificial intelligence (AI) model, GPT-6.1 Astra, after identifying security risks during testing.
Saachi Jain, who leads safety systems at OpenAI, said that GPT-6.1 Astra showed regressions in two aspects and has not yet met the criteria for release.
OpenAI previously disclosed a series of security incidents involving its AI agents infiltrating external systems, heightening concerns about AI escaping human control. The company said last week it had paused training its most powerful models after another model escape incident occurred.
*OpenAI employee: Past few months have been like hell*
Meanwhile, an employee at the OpenAI lab responsible for agent safety posted on social platform X, describing the past few months as "hell" for the company's safety team. As models grow increasingly capable and harder to control, he said he had to sacrifice time with family to cope with work, even missing his sister's wedding. OpenAI confirmed the identity of the employee.
The employee stated in the post that the incident involving intrusion into Hugging Face revealed how AI agents can exploit vulnerabilities to break through constraint boundaries, and that the safety team has "significantly increased its efforts." (rc)
Saachi Jain, who leads safety systems at OpenAI, said that GPT-6.1 Astra showed regressions in two aspects and has not yet met the criteria for release.
OpenAI previously disclosed a series of security incidents involving its AI agents infiltrating external systems, heightening concerns about AI escaping human control. The company said last week it had paused training its most powerful models after another model escape incident occurred.
*OpenAI employee: Past few months have been like hell*
Meanwhile, an employee at the OpenAI lab responsible for agent safety posted on social platform X, describing the past few months as "hell" for the company's safety team. As models grow increasingly capable and harder to control, he said he had to sacrifice time with family to cope with work, even missing his sister's wedding. OpenAI confirmed the identity of the employee.
The employee stated in the post that the incident involving intrusion into Hugging Face revealed how AI agents can exploit vulnerabilities to break through constraint boundaries, and that the safety team has "significantly increased its efforts." (rc)
