Get a local home GPU rig, loadup open source models, write code use opensource SAST to verify code security quality use feedback to train local model further then use frontier models to recheck as a second eye (project code only) - it's the only way you remain safe for now