Unreleased models widen policy blindspots

When AI labs keep high‑capability models and internal evaluations private, outside auditors, policymakers, and the public lose access to the best evidence about near‑term capabilities. That secrecy can make regulators surprised by rapid capability jumps, hinder coordinated safety measures, and encourage strategic withholding that slows public verification. — Information asymmetry from unreleased models creates governance and safety risks by shrinking the public evidence base regulators need to judge and respond to fast AI progress.