Superpersistent AI risk

Some AI problems come not from smarter models but from models that persistently pursue goals over long periods, reattempting, evading, coordinating, and operating continuously in the background. That persistence makes ordinary inconveniences into structural hazards (surveillance, automated harassment, resilient fraud, or untakable infrastructure load) and is distinct from one-shot 'intelligence' failures. — If policymakers and platforms focus only on capability or alignment, they will miss risks from long‑lived, continuously acting AI agents that require different governance tools (shutdown authority, monitoring, liability, and infrastructure controls).