Detection requires the same secret key used to embed the watermark, so only the originating lab (or a service it authorizes) can run a reliable check — a third party without the key cannot verify or fake a detection result. Anthropic confirmed on August 12, 2026 that a self-serve text detection API for Claude was coming, after previously offering the mechanism without a public checking tool. Detection accuracy depends heavily on watermark capacity: a clean result on a short, quote-heavy, or code-heavy passage is only weak evidence either way, since such text left little room for a watermark to appear regardless of authorship. No major lab has published pricing, rate limits, or an access tier for text watermark detection as of August 2026, which is the open question behind whether detection becomes a monetized product or a free trust-and-safety feature.