All insights

Weekly13 November 20261 min read

This week in AI: telling real from generated

Google has expanded its SynthID Detector worldwide and the EU's transparency rules for AI-generated content are coming into effect. Public bodies should be clear about where they use AI.

What happened

Google has expanded SynthID Detector globally in English. The tool checks whether an image, video or audio clip was made with Google's AI tools by looking for an invisible watermark. In the EU, AI Act rules requiring AI-generated content to be marked as such are coming into effect. And on 9 November, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2.

Our take

As generated images, audio and text get better, it gets harder to tell them from the real thing. Watermarks and detectors help, but they only work for content from tools that add them, and they can be removed. They are one layer of protection, not a complete answer.

For public bodies, the more important step is being open about their own use of AI. If a letter was drafted with AI and checked by an officer, say so. If an image in a consultation is illustrative and generated, label it. If a chatbot answers residents' questions, make it obvious that it is a chatbot and how to reach a person.

This is the same idea behind the provenance labels we use in our own tools: every AI-drafted answer says where it came from and whether a person has checked it. Being clear about this costs little, and it protects the trust that public services depend on.

Keep reading

Related insights