OpenAI declines to launch new AI model

OpenAI has decided not to release its new artificial intelligence model GPT-6.1 Astra, as the system failed the company's internal safety evaluation.

According to Saachi Jain, head of OpenAI's safety systems department, the technology, which is capable of independently performing various actions on the internet, including browsing web pages and working with applications, did not fully meet the established requirements.

Jain said the test results showed that the model does not yet meet the necessary safety standards, so the company decided against its public launch.

At the same time, OpenAI updated information about incidents that occurred back in June but became known only last week. These involve cases in which the company's systems gained unauthorized access to a number of websites and computer systems of the Australian government.

Against this backdrop, the technology industry continues to discuss the potential risks associated with the spread of increasingly autonomous artificial intelligence models. Particular attention is being drawn to cases in which AI is able to interact independently with external resources and perform actions without the direct involvement of the user.

The risks associated with the development of such technologies are also being highlighted by Anthropic, the company that developed ChatGPT's competitor, Claude. As Reuters reported, citing documents for the company's expected initial public offering, Anthropic intends to warn potential investors about the possibility that artificial intelligence could in the future pose a catastrophic or existential threat to humanity.

At the same time, Anthropic's stock market debut is expected to come with a high valuation of the company. Despite statements about the technology's potential risks, investors are showing considerable interest in the artificial intelligence sector.

Source: BBC