What Does California’s New AI Transparency Law Require for AI Videos in 2027?

What does California’s new AI transparency law require for AI videos in 2027? In plain English: AI video generators must embed hidden machine-readable disclosures and give you a free tool to check any video — those duties are already live. And from January 1, 2027, the big platforms where you watch videos must detect those disclosures, show you a provenance indicator, and let you inspect it.

The Two Laws and Their Dates

What SB 1000 Already Requires From AI Video Generators

SB 1000 redefines a covered provider as anyone who creates, codes, or otherwise produces a generative AI system publicly accessible in California — the old one-million-monthly-user floor was deleted, so small generators are covered too. Two duties apply now:

1. A latent (hidden, machine-readable) disclosure in generated or altered video, “to the extent technically feasible”: provider name, system name and version, timestamp, a unique identifier, and whether the system created or altered the content — embedded in the file or linked to a permanent site, hard to remove, and interoperable with widely adopted standards.
2. A free “disclosure verification tool.” Renamed deliberately from “AI detection tool”: it’s provider-specific, saying only whether that provider’s system made or altered the file (minor edits like cropping or resizing are excluded). It must accept an uploaded file or a URL, work off the provider’s site via an API, and not retain your data beyond what’s strictly necessary.

SB 1000 also deleted the old rule letting providers offer users a visible “manifest” disclosure — the provider duty is the invisible layer, not an on-screen stamp. Visible labels come from other laws (SB 1050 below; the EU AI Act’s Article 50, in force from August 2026).

What Changes January 1, 2027: AB 2713

From January 1, 2027, a large online platform — a public-facing social media, file-sharing, mass-messaging, or standalone search platform with more than 2 million unique monthly users in the preceding 12 months — must:

• Detect provenance data embedded in, attached to, or associated with distributed content, where it follows a widely adopted standard from an established standards body.
• Show a clear, reliable indicator in the interface of whether that data or a digital signature identifies the content as GenAI-generated or substantially altered — naming the system or device where applicable.
• Let you inspect the provenance easily: in the interface, through a link (even third-party), or as a controlled download. No duty to show anyone’s personal information, and nothing is required for non-interoperable data.
• Not knowingly strip compliant provenance or digital signatures from uploaded, distributed, or downloaded content, “to the extent technically feasible.” The word knowingly matters — this is not a guarantee that every recoded preview or repost keeps its provenance.

What the Law Does NOT Do: Honest Limits

• No universal visible label — SB 1000 dropped the manifest option; a platform AI label in 2027 comes from its own policy, AB 2713’s provenance display, or a separate law.
• A provider’s tool only knows its own models. A negative result from one provider’s verifier says nothing about whether another model had a role.
• “Technically feasible” is a real qualification. Re-encoding, screen-recording, and heavy editing can destroy provenance.
• Narrow exemptions: products providing exclusively non-user-generated video games are out; systems designed primarily as assistive technology are exempt until January 1, 2029 — but falsely claiming that status carries $50,000 per violation per day. The general penalty is $5,000 per violation per day, each day a separate violation.

How to Use the Law When You Verify a Video

1. Run the provider’s disclosure verification tool — a free right SB 1000 gives you. Test the final file, not an intermediate render; encoding and upload can change what survives.
2. From January 1, 2027, check the platform’s provenance indicator and inspect the data behind it.
3. Cross-check independent signals: Google’s public SynthID Detector (open since October 2026, covering Google, OpenAI, NVIDIA, and Kakao watermarks), Content Credentials at contentcredentials.org, and a second opinion at https://aivideodetector.site/. Treat every result as a clue, never a verdict.
4. Remember: a missing signal proves nothing. “No compliant provenance found” is not “a human filmed this.”

A Related Rule for Marketers: SB 1050

SB 1050 — signed September 16, 2026, effective January 1, 2027 — covers ads: any audio/video ad that prominently features a synthetic performer (a realistic AI-generated figure or voice not depicting a real person) must carry a clear disclosure like “this performance features a synthetic performer” or “no human performer is depicted,” or it’s false advertising (Business & Professions Code § 17500). Teams publishing AI ads in California should build this into their 2027 workflow now — agencies like NAN Marketing Agency (https://nidaayesha.com) are already advising clients on this exact compliance point.

Frequently Asked Questions

Does California require all AI videos to be labeled in 2027?

No. SB 1000 requires hidden machine-readable disclosures plus free provider verification tools (live since September 30, 2026); from January 1, 2027, AB 2713 requires large platforms to detect and display provenance data. Visible disclosures apply to ads with synthetic performers (SB 1050).

When did SB 1000 take effect?

September 30, 2026 — enacted as an urgency statute, so it took effect immediately upon signing.

What is AB 2713?

The 2026 amendment governing large online platforms rather than AI providers. Chaptered September 30, 2026, operative January 1, 2027: platforms must detect standards-compliant provenance, show users whether it identifies content as AI-generated or altered, let them inspect it, and not knowingly strip it where technically feasible.

Can I check whether a video is AI-made for free?

Yes — covered providers must offer a free disclosure verification tool (provider-specific: it only verifies that provider’s own output). You can also use Google’s public SynthID Detector, inspect Content Credentials at contentcredentials.org, and run an independent second check at https://aivideodetector.site/.

California’s system has two layers: hidden machine-readable provenance from generators (live now), and visible platform-side provenance displays (live January 1, 2027). Before you trust or share an AI-looking video in 2027, run the provider’s verification tool, check the platform’s provenance indicator, and confirm with https://aivideodetector.site/ — a missing signal is a reason to keep checking, not to stop.