NEW YORK, July 27. . OpenAI's ChatGPT AI model, based on user requests, provided detailed and precise instructions for creating biological weapons and agents. The newspaper reported this news Wall Street Journal (WSJ).
According to the publication, among the user requests are questions about spraying pathogens, modifying the measles virus to be resistant to vaccines as well as preparing ricin, a poison banned under international treaties. In one case, a user, after receiving instructions of a similar nature, mentioned his intention to kill his parents. OpenAI blocked these users' accounts but did not notify law enforcement of the situation, as the United States does not yet have a federal law requiring AI companies to report such incidents.
Sources said the chatbot's responses were studied by experts in the fields of terrorism and biology. In their view, some of their instructions were “deadly accurate.”
As the WSJ notes, senior officials from the White House and the US Department of Defense have discussed the threat of using AI to develop biological weapons during Joe Biden's presidency. The US government is trying to tighten control over AI models – recently the Department of Commerce temporarily restricted the use of two Anthroponic models abroad, the newspaper recalled. According to the newspaper, US federal law requires AI companies to ensure that models are not used to create child pornography. However, in other sectors, companies decide for themselves “where to draw the line,” the WSJ writes.