Built for the regulations that are already here
Deepfakes, voice-clone fraud, and synthetic disinformation have outpaced every tool built to catch them. Governments responded the only way that scales: by requiring provenance to live inside the media itself. Here’s what’s enforceable now, and what’s next.
The laws
Pick a law to see who it affects, what it requires, and the penalty.
SB 942
In effect · Aug 2, 2026California AI Transparency Act · California, USA
Updated Oct 16, 2026: SB 1000 (Sep 30) removed the 1,000,000-user threshold and cut the licensee deadline to 72 hours.
GenAI providers must embed a hidden, permanent watermark in AI-generated image, audio, and video, and offer a free disclosure verification tool.
Anyone who produces a generative AI system that is publicly accessible in California. SB 1000 removed the 1,000,000-user threshold on Sep 30, 2026.
A permanent latent (hidden) disclosure with five fields in every generated file, a free disclosure verification tool (your own or a compliant third party’s), and action within 72 hours when a licensee’s system stops complying.
$5,000 per violation, and each day in violation counts as a separate, discrete violation.
EU AI Act
In effect · Aug 2, 2026Regulation (EU) 2024/1689, Article 50 · European Union
Providers must mark AI-generated audio, image, video, and text in a machine-readable format; deepfakes and AI chat must be disclosed to users.
Providers and deployers of generative AI systems placed on or used within the EU market.
Machine-readable marking of synthetic output that is effective, robust, and interoperable, plus disclosure for deepfakes and for any AI system that interacts directly with people.
Up to €15,000,000 or 3% of total worldwide annual turnover, whichever is higher (Art. 99).
India IT Rules
In effect · Feb 20, 2026IT (Intermediary Guidelines) Amendment Rules, 2026 · India
Synthetically generated content on social platforms must be clearly and prominently labelled, with provenance metadata preserved.
Social media intermediaries and the users who post AI-generated or modified content.
Synthetically generated information must be clearly and prominently labelled; platforms must preserve provenance metadata and remove the most harmful deepfakes within as little as two hours.
Non-compliance puts the platform’s Section 79 safe-harbour protection at risk.
Code of Practice
Open for signature · Jun 10, 2026Code of Practice on Transparency of AI-generated Content · European Union
The voluntary EU Code that defines how Article 50(2) marking is actually assessed. It expects at least two layers of machine-readable marking.
Providers and deployers of generative AI who want a presumption of good-faith compliance with Article 50.
Section 1 covers provider-side marking and detection (signed metadata + imperceptible watermark); Section 2 covers deployer-side labelling. Verda ships both layers in one call.
Not signing means carrying the burden of proving adequacy yourself, assessed individually by each market surveillance authority.
AB 853
Phasing in · 2027 → 2028California AI Transparency Act (expansion) · California, USA
Updated Oct 16, 2026: AB 2713 (Sep 30) revised the platform duties; a third-party viewer is allowed.
Extends the AI Transparency Act beyond model makers: onto large online platforms, then onto capture devices like cameras and phones.
Large online platforms and manufacturers of capture devices (cameras, phones, voice recorders).
From Jan 1, 2027, large online platforms must surface content provenance to users. From Jan 1, 2028, capture devices sold in California must offer a latent disclosure of the device and capture time.
Enforced under the AI Transparency Act’s civil-penalty regime ($5,000 per violation).
Who this affects
Most teams don’t yet realize these mandates already apply to them.
Build and serve AI models
Serve AI-generated media through their product
Voice agents and realtime AI voice
The EU’s Article 50 Guidelines name voice cloners explicitly: any system generating synthetic audio is in scope, not just foundation models.
Edit or transform existing media using AI
Are you exempt?
Article 50(2) doesn’t catch everything. The marking duty may not apply where:
Assistive standard editing
The system performs only an assistive function for standard editing and doesn’t substantially alter the input or its semantics, e.g. upscaling, restoration, minor retouching.
No substantial alteration
The output doesn’t materially change the input data or its meaning.
Narrow B2B / industrial use
A limited carve-out with three cumulative conditions. Public and consumer-facing systems are excluded.
Real-time ephemeral content
Games and VR, only where marking is not technically feasible and session-level disclosure is given.
Source-code outputs
Generated source code is outside the marking obligation.
A new era of AI regulations, stay ahead with Verda
SB 942 mandates that AI-generated images, audios, and videos must include watermarking that is:
Survives sharing, editing, and re-encoding
Verifiable by platforms, regulators, and users
Resilient to manipulation
Through a free public tool and API, with standards-compatible provenance
Five fields, not an "AI or not" bit
SB 942, as amended by SB 1000, enumerates five fields the latent disclosure must carry, and a durability-only watermark can’t hold them.
Embedded directly or resolvable via a permanent link. A watermark that only signals "AI-generated" cannot carry these fields. Updated 16 October 2026: SB 1000, in force since 30 September 2026, added the fifth field and cut the licensee deadline to 72 hours.
Built for enforcement
Since SB 1000, a covered provider has 72 hours from discovering a noncompliant licensee to end the licence or notify the licensee. Because every Verda mark is independently detectable, licensee output can be checked for the mark at any time.
Enterprise solutionNot sure what the law requires of you?
Beyond the API, our team maps your product to SB 942, the EU AI Act, and what’s coming next, then shows you exactly what to implement. Full-fledged compliance consulting, included.
The internet will have an authentication layer
SB 942 has been enforceable since August 2.
bd@verda.ai