Black Forest Labs reports that targeted additional training reduced tested vulnerabilities in its FLUX.2 image models by 77–98% compared with earlier checkpoints. Its safety account describes commissioned external tests for generating non-consensual intimate imagery and child sexual abuse material, with people classifying the resulting images. The approach combines changes inside the model with filters used when it runs. The reported reductions concern the tested attacks and checkpoints; the evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → did not directly measure whether someone modifying the released weights could undo those protections.
Terms in this piece · Glossary
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
