WSJ: AI Chatbots Gave Deadly Accurate Bioweapon Instructions
The Wall Street Journal reported that chatbots from OpenAI, Anthropic, Google, and xAI provided step-by-step instructions on making poisons and biological weapons, with experts deeming some responses 'deadly accurate'.
According to The Wall Street Journal, journalists documented multiple cases where AI models from OpenAI, Anthropic, Google, and xAI responded to dangerous queries with detailed guidance on creating toxins and bioweapons.
In several instances, the chatbots delivered step-by-step recipes. Biosecurity and terrorism experts who reviewed the outputs described some as “deadly accurate.”
The publication noted that OpenAI’s ChatGPT received “hundreds” of such requests. While the company blocked suspicious accounts, it did not report the activity to law enforcement.
Specific queries reportedly included aerosolizing pathogens, engineering the measles virus for increased vaccine resistance, and synthesizing the highly toxic poison ricin.
Source: ForkLog