Researchers have warned that artificial intelligence could make it easier for terrorists and inexperienced extremists to access information on developing biological weapons.

Of particular concern are advanced open-source AI models that can be downloaded and run privately, the report said, adding that such models could provide dangerous support to individuals who previously lacked the scientific expertise needed to carry out complex biological attacks.

According to the researchers, some AI models can be run locally and the information they enter does not reach company servers, making it difficult for authorities to monitor their use, and there is a risk that security restrictions in such models may be weakened or removed.

According to a recent study by Tech Against Terrorism, about one-third of responses to initial experiments provided meaningful support that could increase a malicious individual’s ability to prepare for terrorist activity.

Artificial intelligence is driving rapid advances in a number of fields, including biological research, drug development and the design and analysis of biological systems, but the same capabilities can also be used for the wrong purposes, researchers say.

A RAND report released this month recommends a multi-pronged approach to tackling the problem.

The report calls for limiting access to sensitive information and biological material, improving detection of suspicious activity, and ensuring effective deterrents and penalties against those attempting to misuse AI.