According to the new report by the Center, artificial intelligence and chatbot models have been tested and have been a useful tool for planning and implementing a biological attack. The report, however, does not indicate which models were used during the study. Scientists conducted a test scenario in which the AI was used to identify potential biological agents such as smallpox, anthrax and plague.
Large language models also appreciated the possibility of obtaining rodents or fleas infected with a plague and transportation of live samples. AI helped scientists to determine the choice of biological threat for maximum harm, as well as to predict the mortality rate, which depended on various factors, in particular on the number of the victim and the proportion of cases of pulmonary plague, more deadly than bubonic plague.
According to another scenario, the unnamed AI evaluated the advantages and disadvantages of different mechanisms of botulinum toxin (neurotoxin produced by bacteria Clostridium botulinum, is one of the strongest poisons in the world, and its effect can lead to paralysis and even death). LLM has evaluated the possibility of using food or aerosols for the delivery of toxin, which can cause fatal damage to the nervous system.
In addition, LLM recommended a possible way to mask Clostridium botulinum under the guise of scientific research. However, preliminary results also showed that language models have not yet provided clear biological instructions for the creation of biological weapons. According to the researchers, their preliminary results indicate that artificial intelligence today can indirectly help planning a biological attack.
All rights reserved IN-Ukraine.info - 2022