AI-based chatbots could help plan a biological attack - study

According to a study by US think tank Rand Corporation, artificial intelligence models used in chatbots are capable of providing advice on planning a bio-attack.

Here's What We Know

Scientists have tested several large AI language models. It turned out that they can help prepare a biological attack, although they do not directly generate instructions for creating a weapon.

In one test, an anonymous AI evaluated the potential of different bioagents, such as smallpox and plague, to massively affect people. In another, the model discussed the benefits of botulinum toxin and recommended a legend for acquiring it.

According to the researchers, the results indicate that chatbots can help in planning a bio-attack. However, it needs to be analysed whether the responses are simply a reflection of information already available on the internet.

Nevertheless, AI development companies should limit the models' ability to have such conversations, the researchers said.

Source: The Guardian