’치명적‘ 인공지능이 왔다…오픈AI, 인간 개입 없이 취약점 공격한 ‘아스트라’ 공개

OpenAI has unveiled its latest artificial intelligence model, GPT-6-Astra, declaring it as a "deadly" level of performance in terms of cybersecurity. This new model, which OpenAI describes as an "AGI era" model, is capable of handling a wide range of intellectual tasks. According to the company, Astra is the most intelligent and best-aligned model among all its developed models, performing exceptionally well in areas such as cybersecurity, science, and professional fields.

Notably, Astra demonstrated the ability to perform tasks typically handled by humans through multiple steps with unprecedented speed. For instance, it found all information related to cat care in just 5 minutes and 27 seconds—30 minutes would usually be the time taken by a human. Similarly, job search information that would take a human 5 hours to find was completed in just 2 minutes and 51 seconds.

The model also reviewed 41 financial statements in a few minutes and identified all four deliberately hidden errors. OpenAI emphasized the model's enhanced ability to stay within the authorized limits of its tasks. During the same day, the company announced a $10 billion "Daybreak" program, providing access to the latest model, GPT-6-Astra, to cybersecurity professionals, along with training and technical support.

OpenAI CEO Greg Brockman stated during a press briefing that, in his opinion, we are now in the era of general artificial intelligence. He added, "We can safely say that we are now in the era of general artificial intelligence." However, external evaluations rank Astra behind its competitors. Astra scored 61 points in AI Index, the same score as OpenAI's previous GPT-5.6 model, while Anthropic's top model, Claude 5.1, scored 66 points.