AI models have become highly reliable at writing code that compiles, but they are still failing basic security tests in ...

Veracode tracked 100-plus models for a year. AI-generated code security is stuck at a 56% pass rate, even as AI now writes half of all code.