// MODEL OPTIMIZATION AND PROMPT SYNTAX TERM
Watermarking
Embedding hidden information into digital content, like AI-generated images or text, to prove its origin or authenticity.
TECHNICAL DEFINITION
Watermarking in AI refers to embedding imperceptible or robust digital signatures, metadata, or patterns into AI-generated content (e.g., images, text, audio) to indicate its synthetic origin, track its provenance, or deter misuse, often using cryptographic or steganographic techniques.
BACKGROUND
Generative artificial intelligence (GenAI) is a subfield of artificial intelligence (AI) that uses generative models to generate text, images, videos, audio, software code or other forms of data. These models learn the underlying patterns and structures of their training data, and use them to generate new data in response to input, which often takes the form of natural language prompts.
READ MORE ON WIKIPEDIASYNONYMS & ALIASES
- Digital signature
- Content tagging
- Origin marking
- Authenticity mark
USAGE NOTE
AI watermarking is being explored as a method to distinguish AI-generated content from human-created content.
DEVELOPERS
Organizations developing technology related to Watermarking.
Google develops and implements watermarking technologies for AI-generated content, notably SynthID, which embeds imperceptible digital watermarks into AI-generated images to allow detection.
As a leading developer of generative AI models, OpenAI actively researches and discusses methods for watermarking AI-generated text and images to address issues of provenance and detection.
Microsoft Research conducts extensive studies into AI content authenticity, provenance, and detection mechanisms, including various forms of digital watermarking for AI-generated media.
Adobe is a key player in content authenticity, integrating watermarking and provenance features (Content Credentials) into its Firefly AI-powered creative tools to indicate AI generation or modification.
Co-founded by Adobe, the CAI is a cross-industry initiative dedicated to implementing content provenance standards, including cryptographic watermarking, to identify AI-generated or AI-modified content.
Meta AI researches and develops generative AI models and explores solutions for content authenticity, which includes watermarking and other methods for identifying AI-generated content on its platforms.
As a developer of advanced large language models like Claude, Anthropic is focused on AI safety and trustworthiness, which includes exploring techniques for watermarking or identifying AI-generated text.
IBM Research investigates various aspects of AI trustworthiness, including methods for digital watermarking and provenance tracking of AI-generated data and models.