We use anonymized optimization metadata to improve our algorithms under legitimate interests. Your raw prompts, raw evaluation data, and API keys are not used for aggregate learning. You can object at any time in your Privacy Settings.