OpenAI Announces EU AI Act Watermarking
The Facts
- OpenAI announced on Monday it will add invisible watermarks to eligible ChatGPT and Codex text outputs in the European Union over the coming weeks, a step taken to comply with the EU AI Act's requirement that generated text be machine-readable.
- API customers worldwide can opt in to watermarked outputs for select models starting Monday, though the setting is off by default. OpenAI confirmed it is not making text watermarking a global default at launch.
- The technology, called textGrain, embeds a statistical signal in the model's word choices rather than adding visible characters or symbols. OpenAI plans to release textGrain as open source.
Sources Split
LEFT
CENTER
RIGHT
+1
+1
ESTAB
CRIT
CENTER
PRO
ESTAB
The Spin
Pro-Europe narrative
Knowing whether a human or a machine made what shows up on a screen is the bare minimum for a functioning information ecosystem. Without marking, fraud, impersonation and manipulation at scale run unchecked while people are left guessing instead of deciding. Voluntary promises from AI firms went nowhere for years, with binding rules and real enforcement the only things that actually delivers results.
Anti-Europe narrative
Europe keeps churning out paperwork while American labs chase real breakthroughs, and the result is a compliance circus that proves nothing. A watermark at best only shows if a tool helped generate, not whether the ideas behind it are of value, and at worst can be easily worked around through small semantic changes or use of a different model. This is classic ineffective EU bureaucracy at its finest.
