Bill Gates Raises Alarm Over AI's Potential Role In Bioterrorism

The expanding reach of artificial intelligence (AI), particularly through open-source platforms, could be exploited by non-government groups to advance bioterrorism, Bill Gates has warned. 

Writing in his yearly message titled ‘Optimism with Footnotes’, Gates said AI’s promise to reshape society must be weighed against the dangers it presents worldwide.

“In 2015, I gave a TED talk warning that the world was not ready to handle a pandemic. If we had prepared properly for the Covid pandemic, the amount of human suffering would have been dramatically less. Today, an even greater risk than a naturally caused pandemic is that a non-government group will use open source AI tools to design a bioterrorism weapon,” wrote Gates. 

Gates stressed the need for careful oversight in the development, regulation and use of artificial intelligence, warning that its capabilities could continue to advance without any clear ceiling.

“When people in the AI space predict that AGI or fully humanoid robots will come soon and then those deadlines are missed, it creates the impression that these things will never happen. However, there is no upper limit on how intelligent AIs will get or on how good robots will get, and I believe the advances will not plateau before exceeding human levels,” he wrote. 

Biological weapons rely on pathogens or toxins to spread illness and destruction among humans, livestock and crops. When such methods are pursued by governments, they are classified as biowarfare, while similar actions by non-state groups fall under bioterrorism. The illegal trade in biological agents adds another layer of risk with far-reaching consequences.

Last October, researchers at Microsoft demonstrated how existing safeguards designed to prevent rogue actors from pursuing biological weapons could be circumvented. In a paper published in Science, the team examined 72 regulated proteins, including ricin.

Using artificial intelligence tools for protein design, researchers generated tens of thousands of DNA sequences capable of producing altered versions of regulated proteins, some with potentially harmful properties. These sequences were subsequently submitted to biosecurity screening systems used by DNA synthesis firms to assess whether they would be flagged.

"The tools failed to flag many of these sequences as problematic. Their performance varied widely. One tool flagged just 23 per cent of the sequences," the study said.

According to the authors, DNA suppliers must upgrade their screening technology, and AI firms should embed stronger safeguards within tools used for protein engineering.