☠️ AI Can Create Dangerous Poisons

Microsoft scientists tested whether AI could bypass biosecurity systems when designing new proteins.

The results were alarming—AI successfully bypassed the protections and was able to generate approval for creating toxins dangerous to humans.

⚠️ To produce a new protein, synthetic DNA is typically ordered from a commercial supplier and then introduced into a cell. However, biosecurity systems are supposed to automatically block requests for hazardous compounds.

Using AI models designed for protein creation, typically employed in drug discovery, researchers managed to modify the structure of toxins in a way that bypassed security filters while preserving their lethality.

The experiment was conducted entirely in a digital format, and the specific toxins modeled were not disclosed. However, this method could potentially be used to synthesize untraceable variants, such as ricin (a poison derived from castor beans) or infectious prions that cause mad cow disease.

🛡 The main takeaway from the experiment: existing biosecurity systems need to be updated. Experts suggest tightening verification processes when ordering synthetic DNA. Some researchers, however, believe that filters should be integrated directly into AI models so they cannot generate dangerous molecules in the first place.

#medicine @hiaimediaen