Pinterest has updated the AI transparency disclosure for Navigator to say the models are trained on a combination of synthetic data, public content on the platform and usage data from AI-powered features, according to its official policy page as tracked by AuditSocials. The change was detected on October 13, 2026.
The page previously described Navigator as a single model, Navigator-1, trained only on synthetic conversation data that was not collected from users. It now describes Navigator as a family of multimodal models. Pinterest lists more than 150,000 synthetic conversation examples, more than 1,000,000 image-and-text data points and more than 100,000 conversational-log data points.
According to the disclosure, collected content dates back to Pinterest's 2010 launch, usage data for AI features is collected after October 2026, and collection is ongoing. The page says public content is used under the license granted to Pinterest in its Terms of Service or pursuant to law, and that datasets with personal information are processed under the Pinterest Privacy Policy.
The Pinterest Canvas disclosure also changed, from approximately 500,000,000 rows to more than 600,000,000 data points of public Pins, descriptions and metadata. No effective date is listed. AuditSocials rated the change medium severity.
The update is relevant to marketing, advertising, e-commerce and direct-to-consumer businesses. Advertisers and creators should treat public Pins, descriptions and metadata they publish as content Pinterest may use to train its AI models under the license in its Terms of Service.
Separately, as context only, Pinterest reported 4.72 million moderation decisions to the EU DSA Transparency Database for October 16 to October 22, 2026, up +11.9% from 4.22 million the previous seven days.
