California AI Transparency Act date, and latent vs manifest marks

Per the AB 853 text, existing law made the California AI Transparency Act operative on January 1, 2026, and the bill would delay its operation until August 2, 2026. The text ends with "This chapter shall become operative on August 2, 2026." This post reads the bill text only and does not say whether any particular company is covered; that is a question for a lawyer.

The bill language is from the leginfo page, read 2026-09-30. Sume facts are from the API reference, Timeline compose, Video captions and Video trim docs.

What do latent and manifest mean in the bill?

The bill defines the two terms, and the difference decides which kind of mark you are talking about.

Who has duties and from when?

Per the bill digest, a covered provider must include a latent disclosure in AI-generated image, video or audio content "to the extent it is technically feasible", and offer an AI detection tool. Beginning January 1, 2027 a large online platform must detect provenance data, and beginning January 1, 2028 capture device manufacturers must offer a latent disclosure option. Whether a given service meets the user thresholds is a legal and factual question this post does not answer.

Does Sume embed a latent mark?

The docs I read describe no latent mark on Sume output. A result is a hosted artifact, and the reference says public results should use media.sume.com URLs. Trim is a frame-accurate re-encode by default (libx264, yuv420p), so do not expect file-level data from a source to be carried to the new file. The related post does Sume add C2PA or a watermark covers how to check a file you receive.

Can I add a manifest label with Sume?

A visible label is a manifest-style mark in the bill's vocabulary, and two documented routes can burn one in. Timeline compose takes one still and one video and returns one MP4 with both on screen, so the still can carry the label text. The captions endpoint accepts authored cues (or segments) with text, start and end, which burns your own overlay copy without speech recognition. Whether either satisfies a legal duty is not something these docs claim.