Benchmark measuring how well frontier AI agents complete real long-horizon patient advocacy tasks
BabaBench is Baba's research benchmark for AI agents on patient advocacy. It rebuilds real advocacy tasks, such as housing navigation, insurer calls, paperwork, billing and consent, as simulations that agents must work through end to end with patients, payers, providers and agencies, testing whether an agent can finish the job without privacy violations or prohibited actions. It is available as an arXiv preprint and was accepted to a NeurIPS workshop, per Baba.
For people
For agents
Nothing listed yet.