Transit generates rider graphics and must catch prohibited symbols or unapproved branding before publish. Which policy enforcement should they add?

Brand and prohibited-symbol enforcement is a responsible-AI policy control for generative multimodal outputs. Transit generating rider graphics should flag prohibited symbols and unapproved branding before publish so logos, marks, or disallowed iconography do not ship in customer-facing materials. Detection-plus-block-or-review is the enforcement pattern: generate, scan against brand rules, then release only compliant art. Screening user text prompts for prohibited symbols and unapproved branding before generation fails because it checks inputs rather than outputs, so the model can still invent prohibited symbols or unapproved branding in the generated rider graphics that then ship unreviewed. Flagging prohibited symbols and unapproved branding in training datasets before fine-tuning fails because it cleans source data rather than enforcing policy on the generated rider graphics before publishing, so violating art can still ship. Flagging prohibited symbols and unapproved branding in model feedback reports before publishing fails because it checks internal feedback paperwork rather than enforcing policy on the generated rider graphics before publishing, so violating art can still ship. Exam caveat: generative models may invent plausible but unlicensed marks; similarity—not only exact matches—matters for brand review. Automated flags should escalate to a human brand owner for ambiguous cases. Operational check: maintain a list of prohibited symbols and approved brand assets, run generated rider graphics through the flagging step, quarantine hits for review, and publish only graphics that clear brand policy.