With the help of Claude, humans have foiled several attempts to create biological weapons.

In the early stages of the development of generative artificial intelligence systems, their creators were concerned about safety issues because they believed that some users would try to create new types of weapons, including biological weapons, with their help. Anthropic says it has thwarted several attempts to create biological weapons this year.

Image source: Human

This information is available After the report is released Anthropic on the incorrect use of its artificial intelligence model. At the same time, the company itself emphasized that it cannot be said with certainty that the scientists involved in these incidents intended to create biological weapons. In fact, the field’s queries to AI models are often used to develop vaccines against dangerous pathogens, and it can be difficult to determine the developers’ true intentions based on their ingredients. However, Anthropic decided to bring these incidents to the attention of regulators just as a precaution.

Until now, there has been greater focus on using artificial intelligence models to hunt and exploit vulnerabilities in companies and organizations’ information infrastructure, while biological threats have remained largely hypothetical. Anthropic’s latest report includes “A chilling example of state-sponsored bioweapons creators exploiting rapidly evolving artificial intelligence models.”.

Anthropic’s extensive reporting, covering the past eight months, has traditionally provided examples of representatives of countries unfriendly to the United States using the Crowder family’s AI models to achieve certain political goals. For the first time, examples are given of the use of Cloud to develop traditional types of weapons, including firearms, missiles, bombs, and combat drones.

Anthropic is not as clear-cut in identifying cases that describe attempts to develop biological weapons because it is unsure whether the request’s author had malicious motives. Therefore, the organizations and countries involved in the activities are not named. However, the company was alarmed by the study authors’ attempts to hide their origins and the purpose of such activity, and the corresponding accounts were blocked.

For example, one researcher was trying to study the details of the spread of the chikungunya virus, which is transmitted to humans through the bites of infected mosquitoes and causes long-term joint pain accompanied by severe fever. According to Anthropic, one particular researcher attempted to modify the virus so that it had more obviously harmful properties. In addition to theoretical studies, the researchers also conducted practical experiments on animals, and the laboratories used for these purposes will be provided by a defense research institute.

Just last year, Anthropic itself estimated that its models were not advanced enough to aid humanity in biological research that could lead to weapons creation. Current versions of Anthropic artificial intelligence models are already capable of assisting in such matters, so there is an increasing need to introduce protections and restrictions. Experts emphasize that the current development of artificial intelligence provides opportunities for the creation of drugs and biological weapons, which humans will be unable to resist without preparation. Andrew Webb called for restrictions on the use of such artificial intelligence tools by researchers in countries unfriendly to the United States.