Case Study: Black Forest Labs launches FLUX.2 safely with Cinder Media

Black Forest Labs launches FLUX.2 with Cinder, cutting harmful outputs by >90%

Black Forest Labs faced the challenge of ensuring the safety of its new FLUX.2 generative AI model before its public release. Traditional red-teaming was too slow to keep pace with their iterative development, creating a velocity problem for pre-launch safety testing against high-severity harms like CSAM and NCII generation. They partnered with the vendor Cinder to address this.

Cinder implemented a solution of four iterative red-teaming cycles, testing across text-to-image and image-to-image attack vectors. This approach resulted in a greater than 90% reduction in harmful outputs and made FLUX.2 ten times safer than benchmark industry models at launch. With evaluation turnarounds of under 48 hours, Cinder enabled Black Forest Labs to launch with a verified safety posture and confidence.

Black Forest Labs