This Startup Tests If AI Really Works — Now Backed by $40 Million

AI companies love to brag about how smart their models are. They prove it by running tests called 'benchmarks' — like a standardized exam for software. The problem? Many of these tests are old and public, so companies can train their AI on the answers ahead of time. That's like letting a student see the SAT questions before test day. Vals, a startup founded in 2024, wants to fix that.

Instead of general knowledge quizzes, Vals creates secret tests that measure whether AI can actually do real work — writing a contract, analyzing financial data, or debugging code — at the same quality as a human professional. Because the tests aren't public, companies can't cheat. The startup also checks for dangerous behavior, like whether an AI could help with cyberattacks or bioweapons. Companies pay Vals to run these tests, similar to how students pay to take the SAT.

The 25-year-old founder, Rayan Krishnan, started the company after noticing that AI was advancing faster than the tests designed to measure it. His bet seems to be paying off: Vals now has 25 employees (up from 8), revenue is eight times higher than last year, and investor Andreessen Horowitz just led a $40 million funding round. The company recently launched a program to evaluate AI for federal agencies, too.

So what does this mean for you? As AI gets woven into hiring, lending, healthcare, and legal advice, independent testing becomes a safety check. When a company claims its AI can read X-rays or draft leases, Vals is essentially the referee saying, 'Prove it.' That matters because bad AI doesn't just annoy you — it can deny a loan, misdiagnose an illness, or give dangerous advice. Reliable testing is how we trust AI enough to use it in daily life.