Researchers from a pharmaceutical company wanted to test their artificial intelligence, trained to discover non-toxic drugs, by reversing the parameters. The nightmarish results contained neurotoxins deadlier than any known.
The news has been dominated by artificial intelligence in recent weeks. Far from dystopian predictions like Skynet in Terminator, these AIs mastering writing (chatbots like ChatGPT and Bard), images (Dall-E or Midjourney) or even video (like Gen-1) and are presented as benevolent. However, AI can also be used to create weapons.
Collaborations Pharmaceuticals uses artificial intelligence to try to find new drugs. Four of its researchers took part in a conference organized by the Spiez Laboratory, the Swiss federal institute responsible for protection against nuclear, biological or chemical risks. They showed how their AI, dubbed MegaSyn, could be hijacked to create neurotoxins. They published their finding in the journal Nature Machine Intelligence.
40,000 nerve agents discovered in six hours
Instead of looking for moleculesmolecules having the lowest toxicitytoxicity possible, they asked MegaSyn to look for the highest toxicity and bioactivity. She has been trained with pesticidespesticidesof the toxinstoxins environmental and drug. In less than six hours, the AI generated more than 40,000 results, each more toxic than the next. In particular, she discovered VX, one of the most dangerous nerve agents, invented in 1952 and ten times more deadly than sarin. She also found other Chemical WeaponsChemical Weapons known, as well as many molecules still unknown and potentially more toxic than VX.
The researchers said they had never thought about how their tool could be hijacked before, and were very surprised by the results. With the development of AI which seems to be accelerating, the design of weapons still unimaginable today could quickly be within everyone’s reach…