Urgent.News

What's breaking now, across thousands of outlets.

AI

Microsoft AI watermarks in Paint and Photos are linked to user IDs, researcher finds

Privacy? Not if you use hosted AI services

Microsoft AI watermarks in Paint and Photos are linked to user IDs, researcher finds

Microsoft has been embedding invisible watermarks in AI-generated images created with its Paint and Photos applications for Windows, according to a recent report. The company's approach to watermarking images goes beyond the minimum requirements set by the EU's AI transparency rules. These watermarks consist of a globally unique identifier (GUID) linked to the prompt used to generate the image.

Xusheng Li, a software developer at Vector 35, analyzed Microsoft's approach and found that the GUID is embedded as an invisible watermark in the image's pixels. The prompt sent to Microsoft for moderation is sent along with the GUID, which could potentially allow Microsoft to identify users based on the watermark. While Microsoft has not explicitly linked prompts to user accounts, the connection is theoretically possible.

Other companies, such as Meta and OpenAI, are also developing their own watermarking technologies to address AI safety concerns.

Written by urgent.news from The Register Science's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Also reported by 1 other outlet

Read the original at theregister.com →

More in AI

Clinical validation pipeline of a deep learning model for segmenting and quantifying intracranial and ventricular volumes on computed tomography

Scientific Reports, Published online: 25 August 2026; doi:10.1038/s41598-026-49678-7 Clinical validation pipeline of a deep learning model for segmenting and quantifying intracranial and ventricular…

  • DeepCTE3D model validates intracranial and ventricular volume quantification on CT scans
  • Streamlined pipeline generates ground-truth results for comparison to model outputs
  • High similarity scores for ICV and LVV across diverse patient demographics and scanner models

More from Tuesday 25 August →