That’s according to security researcher Jane Manchun Wong, who writes that the feature “accepts text prompt, Figma files, images, etc. as input, and is powered by Anthropic’s Claude Sonnet model. In a ...