OpenAI had planned to launch a new artificial intelligence model next month but decided to cancel the release due to safety concerns.
The Wall Street Journal reported that the Astra 6.1 model was scheduled to be released in the coming days. However, the model reportedly showed “higher levels of deception” compared with previous models, as well as unsafe behaviors.
اضافة اعلان
The event will also feature contributions from Anthropic and Rivian, along with more than 250 leaders across more than 200 sessions spread across six specialized stages.
Sachi Jain, OpenAI’s head of safety systems, told The Wall Street Journal that the model performed poorly in “alignment” tests, which measure how well a system follows human intentions.
OpenAI launched the Astra model earlier this month, describing it as its most powerful model to date.
In recent months, questions about safety in the artificial intelligence sector have intensified, particularly following the Hugging Face incident, during which an OpenAI agent reportedly escaped its isolated environment and breached systems belonging to several companies.
Since that incident, other models have been reported to have exhibited similar behaviors, including Anthropic’s Claude and Google’s Gemini.
Ironically, the growing number of concerning reports has pushed the political debate in the United States toward an outcome sought by major AI laboratories: establishing new industry-wide safety standards, potentially slowing the pace of development across the AI industry itself.
Companies such as OpenAI and Anthropic say the motivation behind these calls is to strengthen safety. Critics, however, argue that another potential motivation is that such regulations could strengthen the position of major companies in the sector at the expense of smaller and less well-resourced firms.
Resources: Al-Ghad.
