AI Misses Chemical Threats—Here’s Why That’s Dangerous

AI tools are increasingly used to assess risks in science and security, but a big blind spot might be hiding in plain sight: chemistry. Right now, most AI safety evaluations focus on biological threats like viruses or bacteria, assuming that chemistry risks—like toxic gases, explosives, or hazardous materials—are covered under the same umbrella. That’s a problem, according to experts who say chemistry is entirely different and needs its own set of safety checks.

Chemistry and biology may seem similar because both involve experiments with lab equipment like microscopes and mass spectrometers. But the risks they pose are not the same. A dangerous chemical might be overlooked by AI systems trained mainly on biological threats, especially if the model doesn’t understand how chemicals behave under different conditions. Think of it like a doctor who’s great at diagnosing infections but might miss a poisoning case because they’re not trained to recognize chemical toxins.

This matters because AI is being used more and more to monitor labs, research, and even security threats. If AI systems don’t have the right training to flag chemical dangers, harmful substances could go unnoticed—whether in a factory, a research lab, or even a homemade experiment. Experts with chemistry backgrounds are now pushing for AI safety evaluations to include specific chemistry risk assessments, not just biological ones.

The catch? Most AI evaluators aren’t chemists, and the systems aren’t built to spot chemical hazards the way they spot biological ones. Until that changes, we might be leaving a big gap in safety where toxic chemicals could slip through.