tl;dr they tested GitHub Copilot against 89 risky coding scenarios and found about 40% of the roughly 1,700 sample implementations Copilot generated in the test

tl;dr they tested GitHub Copilot against 89 risky coding scenarios and found about 40% of the roughly 1,700 sample implementations Copilot generated in the test were vulnerable (which makes sense given it's trained on public GitHub repos, many of which contain sample code that's a nightmare from a security perspective).