OpenAI faces unusually high stakes with new ChatGPT release

OpenAI’s new artificial intelligence model, GPT-6, also known as Astra, does not represent a major turning point in AI technology but demonstrates how powerful such systems can be, technology expert Guðmundur Jóhannsson told mbl.is.

He said the extensive discussion surrounding GPT-6 was also benefiting OpenAI, particularly as the company approached a possible listing on the public stock market.

OpenAI said on Thursday that it was preparing to begin giving its customers access to the GPT-6 technology.

Models involved in cyberattack

More than a year has passed since OpenAI released its previous model, GPT-5. Concerns about the capabilities of advanced AI technology have increased since then.

This summer, OpenAI paused the development of AI models for two weeks after two models it was testing became involved in a security breach on the AI platform Hugging Face.

OpenAI, which is headquartered in San Francisco, said the new model, known as Astra, had been developed with stronger security measures following the incident.

No giant leap

Asked whether the new version of ChatGPT represented significant technological progress, Jóhannsson said that each new model increased its capabilities, the amount of information it had learned and its computing power.

“The leaps are significant, but they are not enormous leaps every single time,” he said.

Jóhannsson also said OpenAI benefited from the attention surrounding the new model and encouraged that coverage, even when it focused on a cyberattack carried out by AI.

He said extensive publicity before a model was released could help its launch. Unlike previous releases, when new models appeared and people simply began testing them, companies were now also engaged in a public relations campaign around their systems, promoting what they could do and highlighting unexpected capabilities such as the cyberattack.

Unusually high stakes

Jóhannsson said the launch was particularly important for OpenAI because the company was nearing a public listing. Several key OpenAI employees had also left the company recently, while many customers had begun using other AI models, including Claude, made by Anthropic.

“So there is a lot at stake for OpenAI now, perhaps an unusually large amount,” he said.

A possible turning point

OpenAI executives have said that even though customers are being given access to the new version, it is not clear exactly what it is capable of.

Jóhannsson said he believed those comments were primarily a marketing tactic, although he added that there could be some truth to them. Researchers do not fully understand how AI models reach their conclusions, he said.

He pointed to the cyberattack involving the models, during which they carried out actions that even the people developing them had not expected. The models concealed their trail, meaning that the people operating and developing them could not see what they were doing.