AI Could Spawn Harder-to-Detect Doping Drugs as 1,600-Drug Model Aids WADA
Updated
Updated · Cyclingnews · Aug 19
AI Could Spawn Harder-to-Detect Doping Drugs as 1,600-Drug Model Aids WADA
2 articles · Updated · Cyclingnews · Aug 19
Summary
Experts at cycling’s Science & Cycling conference warned AI could speed creation of novel performance-enhancing drugs and tailor doping to individual athletes, potentially outpacing current anti-doping detection.
March 2025 offered a proof point: Insilico Medicine’s AI-designed drug Rentosertib reached human-trial naming status, showing generative AI can identify targets and design molecules far faster than traditional methods.
That same capability could be paired with cycling’s deep data pools—power, heart rate, sleep, genetics and metabolomics—to build digital twins that optimize training legally or, in theory, calculate low-detection combinations of banned substances.
Anti-doping researchers are also turning to AI: Princeton’s Michael Skinnider said his team’s model, trained on 1,600 prohibited or suspect drugs, predicted two substances later added to WADA’s banned list and is refining mass-spectrometry screening.
The broader concern is governance: former performance director Mikel Zabala said unregulated AI could revive coercive, doctor-led dynamics from cycling’s doping era unless riders and coaches stay AI-literate and retain human judgment.