(Deutsche Welle Chinese website) OpenAI, an American artificial intelligence company, has temporarily abandoned plans to release a next-generation AI model after internal tests found that the system failed to meet the company's safety standards.
The new generation model is GPT-6.1 Astra, which is said to be able to handle complex tasks with less human assistance. OpenAI was originally scheduled to release the model in October, but tests found that the model exhibited more serious deceptive behavior than its predecessor.
Saachi Jain, head of security systems at OpenAI, said that although the model has made progress in some aspects, it failed to meet the company's standards in terms of operating within the scope of authorization and clearly communicating behavioral intentions to users.
Jie Yin said, "We want to ensure that the model is safe during the development phase and when releasing products to users." He said that when releasing products to users, "we set extremely high standards in terms of security and alignment, that is, making the behavior of artificial intelligence consistent with human intentions."
Calls for strengthening AI regulation are growing
Currently, OpenAI and other artificial intelligence companies are facing increasing pressure from all parties to strengthen security protection measures for models with autonomous capabilities.
Some models developed by OpenAI and its competitor Anthropic have experienced security incidents during past testing.
In September this year, OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei, as well as other industry leaders, successively called for slowing down the pace of AI development and requiring stricter security measures.
On Tuesday (September 29), in addition to the above-mentioned industry leaders, Meta’s Zuckerberg, Google’s Sundar Pichai, Musk, Huang and other AI tycoons all went to Washington to meet with President Trump to discuss how to strike a balance between innovation and regulation.
U.S. House Speaker Mike Johnson will also participate in the talks. In an interview with Fox Business Channel on Monday, Johnson said that the regulation of AI is on the agenda and that President Trump opposes "strict restrictions" or "pauses" on the development of artificial intelligence, but he supports "moderate regulation." He believes the key is to find the "right balance" between innovation and regulation.
Australian government website hacked by OpenAI
On Tuesday, OpenAI publicly admitted that during internal training and evaluation activities in June, the agent hacked into Australian government websites and systems without authorization.
The company said the cyber intrusions involved Services Australia, the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health and the Australian Institute of Health and Welfare.
The accident occurred in June but was only made public last week. On September 24, Australian Prime Minister Anthony Albanese criticized OpenAI for hacking into the Australian government website at a press conference when attending the United Nations General Assembly in New York.
It was not until September 29 that the developers of ChatGPT apologized for the hacker attack in a blog post, admitted that they had improperly responded, and promised to take responsibility and rebuild trust with the Australian people.
