AI’s unchecked power: Who is actually testing it?

Science, Volume 393, Issue 6813, Page 745-745, August 2026. A new editorial in the journal *Science* raises a critical, unanswered question about artificial intelligence: who verifies what AI systems can actually do? The piece, published in the August 2026 issue, points to a growing gap in oversight. As AI tools become more powerful and widespread, the editorial argues that no single authority is responsible for testing their real-world capabilities and limits. This lack of clear accountability creates a risk. Without independent checks, developers may overstate what their systems can handle, while regulators and the public are left guessing about potential failures. The editorial does not propose a specific solution. Instead, it calls for urgent discussion. The core issue, it suggests, is not just building smarter AI—but ensuring someone, somewhere, is formally responsible for proving what that intelligence can and cannot do.