Anthropic is opening access to a watermark-verification API for Claude-generated text, according to The Decoder. The tool is designed to let approved outside organizations check whether a piece of text contains a digital watermark associated with Claude output. The initial access list is selective. The Decoder reports that regulators, law enforcement, media organizations, fact-checkers, independent researchers, educational institutions, EU civil society groups, and enterprises with their own compliance needs can request access. Anthropic reportedly plans to expand availability over time. The move is tied to watermarking obligations under the European Union’s AI Act. The Decoder reports that since August 2, 2025, the law has required new Claude models to place invisible watermarks in text output. On the available evidence, the verification API is not a general public detector; it is a controlled channel for checking whether Claude’s watermark is present. Technically, The Decoder says Anthropic’s system builds on Google’s SynthID text method. The approach changes word-selection randomness in a way that creates a statistical pattern, rather than appending visible labels or metadata. Anthropic says the pattern may remain detectable after some editing, which would make it different from conventional AI-text detectors that try to infer authorship from style alone. Anthropic’s position, as reported by The Decoder, is that the watermark does not include user data and does not change the quality or substance of generated text. That is the central operational claim for customers and regulators: the company is presenting watermarking as a provenance layer, not as a content-retention or surveillance mechanism. The trade-off remains disputed. The Decoder notes that critics argue watermarking can affect output if a model is pushed toward words chosen partly for a watermark key rather than meaning. It also cites transparency concerns raised by Artificial Lawyer, including situations where detectable AI use could matter in contracts that restrict AI assistance or in disputes over professional fees. For now, the practical change is narrower than a public “AI detector.” Anthropic is giving selected institutions a way to verify Claude watermarks, while reserving broader access for later. That makes the first phase as much about governance and trust relationships as it is about detection accuracy. Who benefits: Regulators, fact-checkers, media organizations, researchers, and enterprises with compliance obligations get a formal path to verify Claude watermarks. Anthropic also benefits if controlled verification helps it satisfy legal and trust requirements without making the detector fully public. Who's exposed: Claude users in settings where AI assistance is restricted could face new detectability if watermarks are checked by approved parties. Organizations adopting the API may also need policies for false negatives, edited text, and how verification results are used.