Can AI Get Immigration Right?

Immigration Bench, developed by the Talent Mobility Fund and Vals AI, evaluated 25 AI models on 262 private test questions drawn from an expert-created set of 519 U.S. immigration-law scenarios involving students, employers, entrepreneurs, academics, self-petitioners, university advisors, and people with DACA, TPS, or other temporary statuses. The leading model, Claude Fable 5.1, met 75.4% of rubric requirements on average but provided fully complete answers on only 26.3% of questions, while performance was especially weak for University Advisor and Fragile/Temporary Status scenarios. Models generally explained established legal rules more effectively than they applied those rules to an individual’s circumstances, and asking follow-up questions produced little overall improvement. Recent 2025–2026 policy changes were the largest gap: models met only 7.3% of related criteria without web search, and search improved results only partially, despite being most helpful for employer questions. The benchmark concludes that AI can assist with general immigration education, research, and drafting but should not be relied upon for individualized legal guidance, particularly in fast-changing or high-stakes cases, and calls for retrieval-grounded tools developed and tested with immigration attorneys, legal organizations, and government experts.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.