Using AI to respond to your emails? You might want to think again.
Previously, AI detectors relied on detecting patterns in writing – mechanical, precise, too “perfect.” We humans have also become better at detecting AI. AI prose generates the same Uncanny Valley effect in both text and graphics – subtle cues that generate the “ick.” However, the high degree of false positives made findings suspect.
As of August 2nd, that is changing. Article 50 of the EU AI Act requires providers of AI systems that generate synthetic content add machine-readable markings to their outputs. This is not the obvious em dashes or zero-width characters; these are verifiable provenance signals. Anthropic and Google chose watermarking – a statistical fingerprint that is completely invisible, probabilistic and impartial.
How does watermarking work? Think of predictive text. If you typed in “Hi, how are…” there is a pretty high chance the next word will be “you.” Similarly, an AI watermark gently nudges word choices that cumulatively create a detectable statistical fingerprint. That signal gets stronger with length but remains undetectable to the reader – like invisible ink added to your prose.
Nearly 200 companies signed on to the voluntary Code of Practice on Transparency of AI-generated Content by the end of July, including Microsoft, Google, Meta and OpenAI (although execution will vary by company). Anthropic launched watermarking worldwide in all Claude models launched after August 2nd.
So how will this affect pharma? The watermarking itself may matter less than how AI was used. FDA and EMA issued joint principles for good AI practice back in January, emphasizing defined context of use, human oversight, traceability and risk-proportionate controls. Neither mentions watermarking or content marking.
I think the greatest hit will be to personal credibility. Would you hire a person whose cover letter was entirely written by AI? Could email platforms eventually use provenance signals as a measure of trust?
How do you think this will affect you and the pharma industry?