Anthropic to Watermark Claude Text Under EU Mandate
· Technology · The Guardian, Indian Express
Anthropic announced it will modify its Claude AI model to embed a watermark in every text output, as required by a new EU regulation that takes effect in December. The watermark appears at a granular, random level and will be invisible to most readers, while Anthropic maintains that it will not impact the model’s overall quality. Tech commentator John Gruber argues the watermark will constrain word choice and make the model’s prose less precise, whereas Professor Steven Murdoch of University College London believes the change will have no noticeable effect. The model will still rely on stochastic selection of words such as “grey” or “overcast” for day descriptions, a feature Anthropic says the watermark will not alter for the public. Critics warn that limiting randomness could degrade the naturalness of Claude’s responses, but Anthropic has not yet supplied evidence of a quality drop.
Why it matters
The EU rule forces any AI-generated text used within the EU to carry a watermark, impacting developers and users of Claude who must comply. The change could alter how consumers and businesses receive AI content and may influence the broader adoption of language models in regulated markets.
Read the original report — The Guardian
Join us on Telegram
Breaking news the moment it lands. At 10,000 members we ship the Android app.