Bugcrowd Introduces Reinforcement Learning Environments for AI Security Training

Bugcrowd announced in a press release the launch of Reinforcement Learning (RL) Environments, a new product designed to train AI models on real-world security tasks. The platform allows AI systems to find, exploit, and fix genuine software vulnerabilities using real code rather than synthetic data.

The RL Environments are built on technology from Bugcrowd’s acquisition of Mayhem Security and are already being used by major large language model developers. Each training environment uses authentic open source vulnerabilities and provides objective scoring for every stage of the process, from detection to remediation.

Bugcrowd stated that the platform helps AI developers accelerate model training by providing ready-to-use infrastructure, removing the need to build complex security environments from scratch. The environments include hundreds of thousands of training scenarios and are designed to teach models both offensive and defensive security skills.

According to the company, all training data comes exclusively from open source software, with no customer or researcher data involved. The offering is aimed at frontier AI teams and large language model providers seeking to develop agents capable of real-world security reasoning.