Vals raised $40 million led by Andreessen Horowitz on 19 September to scale its confidential AI model evaluation framework — structured as a proprietary benchmarking service in which test materials are never published, preventing the exam-gaming that has compromised all major public AI benchmarks. The company evaluates models across law, finance, coding, cybersecurity, biosecurity, and mental health with task-specific test sets that clients cannot train against. Revenue is up 8x year-on-year; the team has tripled to 25. Co-founder Rayan Krishnan describes the business model as "companies paying to take the SAT" — and the company recently launched a federal agency evaluation track. As Anthropic and OpenAI approach IPOs and AI regulation tightens, third-party independent model evaluation infrastructure is becoming a compliance asset. Vals is positioning to be the authority that certifies model safety and capability claims for the institutional and regulatory audiences that will require independent verification.