OpenAI is adding text watermarking in ChatGPT and Codex
News Source
β’2026-10-05T14:08:39-04:00
π° What Happened
OpenAI is adding an invisible watermark to text from ChatGPT and Codex. The mark is machine-readable, so software can spot words written by AI.
It is rolling out first to users in the European Union. OpenAI says it will not yet be the default everywhere.
OpenAI says its system, called textGrain, matched or beat other methods. These include Google DeepMind's SynthID, which Anthropic also used for its own watermarking in August.
π The Backstory
The EU's AI Act is a new law that sets rules for AI tools. It pushes companies to label content made by AI.
Watermarks can help spot fake news and cheating. But some users were unhappy when Anthropic added them without much warning.
Tech firms are under pressure to make clear what was written by a human and what was made by a machine.
π― Why It Matters
As AI text spreads online, it gets harder to trust what you read. Watermarks could help people and schools tell real writing from machine output.
OpenAI is adding an invisible watermark to text from ChatGPT and Codex. The mark is machine-readable, so software can spot words written by AI.
It is rolling out first to users in the European Union. OpenAI says it will not yet be the default everywhere.
OpenAI says its system, called textGrain, matched or beat other methods. These include Google DeepMind's SynthID, which Anthropic also used for its own watermarking in August.
The EU's AI Act is a new law that sets rules for AI tools. It pushes companies to label content made by AI.
Watermarks can help spot fake news and cheating. But some users were unhappy when Anthropic added them without much warning.
Tech firms are under pressure to make clear what was written by a human and what was made by a machine.
As AI text spreads online, it gets harder to trust what you read. Watermarks could help people and schools tell real writing from machine output.