Path 06 of 07 · Researchers and research staff · intermediate

AI with research data

By the end: You can match a dataset's risk classification to an approved Stanford AI service, have requested AI API Gateway access if your data needs it, and have produced one cited literature scan you verified source by source.

The sequence

External Responsible AI at Stanford What counts as High Risk in research terms: participant data, health information, CUI and ITAR material, proprietary code, anything under NDA. Only Stanford-covered services may see it. Open Guide Know Your Risks: Agentic AI Risk Classifications | University IT Stanford's AI risk tiers name research explicitly: literature search assistants are Low; participant screening and research where human oversight is essential are High. Classify your study before choosing a tool. Open Guide AI API Gateway | University IT For pipelines, not chat: the same models by API, approved for High Risk data and PHI, billed to a PTA. Request a key here if scripts will touch your data. Open KB article KB00019978: Stanford AI API Gateway Access and Usage What the key unlocks: /v1/chat/completions for extraction and classification, /v1/embeddings for similarity search over a corpus, the current model IDs and the parameters. SUNet login required. Open Course Claude Platform 101 intermediate1h 23mClaude Academy Anthropic's developer course, from single API requests to agent loops, with runnable TypeScript. The gateway speaks the OpenAI-style chat format from the KB article, so map the concepts, not the SDK calls. Open KB article KB00020852: How to run cited research in Gemini Enterprise Apply it: Gemini Enterprise is approved for PHI, and its Deep Research agent turns one question into a cited brief. Click every citation before you trust the brief. The Standard tier allows one Deep Research run per day. Open