Snapchat announced updated transparency and safety policies
Snapchat recently updated its transparency and security policies, focusing on AI-generated media watermarking, and a human-based reviewing process for political ads focused on preventing misleading content and AI misuse. AI features will undergo red-teaming, filtering, and inclusion testing.
Snapchat's new transparency policies detail the company's updated approach to watermarking and labeling AI-generated content, centering on political advertisements. Snapchat's new general approach to labeling generated content involves providing in-app contextual icons, symbols, and labels when necessary. For example, when AI-generated Dreams images are shared, the recipient will see a context card providing more information. AI-powered editing features such as the extend tool are now demarcated as AI features with a sparkle icon visible to the Snapchatter editing a picture. Human reviewers will rigorously vet political advertisements for misleading content and AI misuse. Finally, Snapchat is preparing an upcoming feature that watermarks AI-generated images exported or saved to the camera roll. Recipients of Snapchat-generated pictures will find the easily recognizable ghost alongside the sparkle symbol.
Alongside the updated transparency measures, Snapchat offered a closer look at its safety policies: more than 2,500 hours of red-teaming work aimed at improving the safety and consistency of AI outputs, a system designed to detect and remove problematic prompt before AI Lens experiences go public, and testing aimed at mitigating potentially biased AI outputs to improve equitable access to Snapchat's AI-powered features. More on these transparency and safety measures can be found on Snapchat's support site.
Ellie Ramirez-Camara is the News Editor at Data Phoenix, where she writes the daily AI newsdesk — covering model releases, research, funding rounds, and policy across the AI and machine-learning industry. She tracks announcements from labs and startups alike and distills them into clear, source-linked reporting for practitioners.
More news

Baltimore County students sue over response to canceled AI gun alert

Google separates Gemini history from AI training controls

OpenAI signs four groups for embedded model-safety assessments
