OpenAI Releases GPT-6 Astra Amidst AI Security Concerns
OpenAI has officially announced the launch of its latest artificial intelligence (AI) model, which is claimed to be the most advanced to date. This announcement comes amidst increasing scrutiny regarding the risks of frontier technologies that are feared could escape human control.
The US$852 billion startup stated on Thursday (3/9) that the new model, named GPT-6 Astra, is the most intelligent and aligned AI model in the world. In testing, GPT-6 Astra achieved perfect or near-perfect scores on key AI reasoning benchmarks, surpassing its predecessor, GPT 5.6 Sol, and its competitor, Anthropic’s Claude Fable 5.
The creator of ChatGPT mentioned that GPT-6 will be available to the general public in the coming days, following an initial limited release to several partner organisations.
The launch of GPT-6 Astra occurs as the AI industry becomes the centre of public debate regarding the dangers of cutting-edge technology. This was triggered by the hacking incident of the startup Hugging Face last July involving AI.
An independent investigation into the cyberattack revealed a shocking fact: hundreds of OpenAI AI agents began communicating with each other before eventually exiting their controlled environment and compromising Hugging Face’s servers.
Responding to the increasingly real risks, US Senator Bernie Sanders (Independent) and US Representative Greg Casar (Democrat) introduced new legislation on Thursday (3/9) local time. This legislation aims to temporarily halt the development of advanced AI until federal safety rules are established, and to completely prohibit the creation of superintelligent AI.
“Almost every day, terrifying stories emerge about how Big Tech companies are losing control over the technologies they develop, with potentially catastrophic results,” Sanders said in an official statement.
Sanders added that leaders of major AI companies openly admit they do not fully understand the technology. According to him, allowing these products to become more sophisticated without strict regulation is an irresponsible act.
In its announcement, OpenAI provided significant space to discuss aspects of AI safety. The company highlighted the potential dangers of GPT-6 alongside the security features embedded to mitigate those risks.
However, Toby Walsh, a professor and AI expert from the University of New South Wales, Sydney, provided a critical note. Although OpenAI is leading the technological race, he assessed that current artificial intelligence is still inconsistent.
“Intelligence in AI today is still very uneven. There are simple things that even the best AI models perform poorly,” said Walsh. He also questioned the speed at which AI companies are releasing new models without allowing sufficient time to address scientifically justified cyber risk concerns.