AI technology can create 40,000 chemical weapons compounds in just about 6 hours

An artificial intelligence (AI) machine learning model generated more than 40,000 virtual molecules that could be used to create chemical weapons in less than 6 hours.

In a paper published in the journal Nature Machine Intelligence, Collaborations Pharmaceuticals describes how the MegaSyn machine learning model is used to create chemical weapon molecules.

This model is often used to prepare potential disease-modifying drugs, which work by removing toxic molecules before scientists synthesize them. Instead of removing these molecules, however, Collaboration Pharmaceuticals reversed the process by adding a pre-existing toxicity database and instructing the AI ​​to 'reward' toxicity. As a result, the AI ​​created some deadly compounds that weren't in the original data. Most of these compounds are toxic.

Fabio Urbina, lead author of the paper and senior scientist at Collaborations Pharmaceuticals, said that his company did not want or plan to synthesize these compounds to test the effectiveness of the experiment. According to him, the main purpose of testing is to find solutions to problems using AI machine learning techniques.

Mr. Urbina explained that the bad guy may need only one toxicity database, and many publicly available databases to create these molecules. They will only need an open source machine learning algorithm, Python coding, and machine learning knowledge to recover the piece of data.

Despite the potential threat, Mr. Urbina is not arguing about stopping open source knowledge sharing or the need to tightly regulate the AI ​​market. Instead, he says, tracking who is using data when it's potentially dangerous is seen as an effective move. He also suggested that machine learning researchers should be trained in the potential dangers of misuse, just like chemistry students.