DeepMind is turning production data into intelligence about the safety of deployed AI models. Safety Oversight uses large-scale traffic and automated evaluation to monitor safety and alignment, measuring real-world efficacy of our safety stack-from in-model training to out-of-model mitigations-for safe models used for broad public benefit.
The GenAI Safety team works with Gemini and GenMedia to detect and mitigate production issues and user misuse at scale in a fast-paced, highly collaborative