The trust and compliance layer for media on the internet
EU AI Act Article 50 and California SB 942 require AI-generated media to carry durable, machine-readable marks. Verda embeds them at generation, in three lines of code, and verifies them free anywhere. Existing EU systems have until 2 December 2026; SB 942 has been enforceable since 2 August 2026.
3 lines of code to integrate. Up to 3% of global turnover if you don’t.
Most teams don’t yet realise these mandates already apply to them. Both laws are enforceable now, and the fines compound.
Each day in violation counts as a separate, discrete violation.
Whichever is higher.
Updated 16 October 2026. California SB 1000, signed on 30 September 2026 and in force immediately, removed SB 942’s 1,000,000-user threshold, renamed the AI detection tool the disclosure verification tool and allowed a third-party one, added a fifth field to the latent disclosure, and cut the licensee deadline to 72 hours.
Compliance as a service
Embed a durable watermark, register it in a central provenance registry, and verify any file anywhere. Three API calls, not a six-month project.
- Sign at creationEmbed
- Record the originRegister
- Verify anywhereTrace
- Built for enforcement72-hour rule
A unique identifier is embedded into the media at creation: imperceptible to humans, resilient to modification.
Verda assigns and stores the registry record: owner, timestamp, integration, file hash, and any metadata you attach.
Submit any suspect file. Verda extracts the watermark and returns the full provenance chain: owner, time, and method.
Since SB 1000, covered providers have 72 hours to revoke or notify a licensee whose system stops complying. Because every Verda mark is independently detectable, licensee output can be checked for the mark at any time.
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 sixth from 2029. A durability-only watermark can’t hold them.
Updated 28 October 2026. These are now dedicated API fields. Every watermark resolves to a permanent verification page that shows them, and each one is signed into the content's certificate. Ask for the California profile and an encode missing one is refused before it is charged.
What you send us with each file
Tell us what the AI did. We record it against the watermark, sign it into the certificate and show it on the content's verification page.
The disclosure
Whether the AI created or altered the content, the AI system's name and version, and when it was made. The provider is always your company's legal name: we verify your incorporation details once, then fill it in on every encode. Nobody can sign under a name they typed.
A law profile
Ask for us_ca, eu_art50 or both, and an encode missing a fact that law needs is refused, naming the field, before any credit is taken. Leave it out and nothing is enforced.
Optional details
Licence, AI-training permission, whether it shows a real person, editorial review, the tools and steps used, and your own fields. Each one is public or private. Prompts and your internal references stay private unless you choose otherwise.
Proof you can keep
The certificate signs the public facts and a fingerprint of the private ones. Months later you can prove exactly what you disclosed, and when, without having shown the private details to anyone.
Updated 28 October 2026. EU coverage: the Code of Practice asks for signed metadata in the file as well as a watermark. Verda gives you the watermark and a signed, timestamped record it resolves to; C2PA manifests in the file are coming.
Metadata isn’t durable. Watermarks aren’t identity.
C2PA metadata gets stripped on re-upload, and blue checkmarks don’t travel with the file. Verda’s watermark lives inside the media and resolves to a signed, timestamped record, so the proof survives when the metadata doesn’t. C2PA manifest writing is next on our roadmap.
Built to hold up in production
Survives the everyday
Compression. Cropping. Re-encoding. Screenshots. Format conversion. Inside the media, not alongside it: tested across the transformations the open internet throws at it.
Built for C2PA
Verda is on the C2PA soft-binding algorithm list, and manifest writing is coming. Our durable watermark is the resilient layer that persists when metadata gets stripped.
Provider-resolvable
Every watermark resolves to the provider that made it, with a full provenance chain from creation to detection. Verified creator identity is an optional layer on top: the trust hierarchy behind "Verified by Verda."
Production-ready
Designed for production pipelines. Minimal overhead: light enough to co-locate with any model, at any scale.
One watermark across image, audio, and video
The same identifier, embedded natively in each medium’s own signal, imperceptible in all three, and verified by one registry. Scope today is image, audio and video; text marking is on the roadmap.
Monet
A signature woven across every pixel: imperceptible, fast, and available as tensor-level integration.
Bach
A multi-layer payload embedded across the waveform: sub-second encoding, resilient to compression.
Pompeii
Per-frame embedding that runs in-pipeline: faster and cleaner than file-based approaches.
Deploy your way
Same SDK, same model, same registry. Choose the tier that fits your privacy, performance, and infrastructure requirements.
Cloud-hosted watermarking
Send media via REST API. Verda handles the GPU, the model, and the registry. You get a watermarked file and a provenance record.
- REST + Python SDK
- Pay per operation, no commitment
- Auto-scaling, zero infrastructure
- Full registry access and decode
Three lines of code
One interface for cloud, managed GPU, and on-prem. Switch tiers by changing a single parameter.
from verda import VerdaClient client = VerdaClient(api_key="sk-...") result = client.encode( file="song.wav", content_type="audio", ) print(result.verify_url, result.download_url)
Who it’s for
SB 942 applies now to any GenAI provider publicly accessible in California, at any size; the EU marking obligation binds systems already on the market by Dec 2, 2026. Dual-track: embed at the model layer, expose a public verification API (SB 1000 also allows a compliant third-party one).
The Article 50 Guidelines name voice cloners explicitly. The realtime "ephemeral" exemption applies only where content is not recorded and marking is not feasible. Most call platforms fail both.
Detect provenance across upstream watermarking schemes and surface authenticity in your UI. One integration for cross-provider verification.
Protect licensed content and trace leaks back to source. Identity-bound provenance for every asset in your catalogue.
Regulatory compliance and internal provenance, media kept in-network. A durable watermark layer, built to sit alongside C2PA manifests.
Classified content tracking in fully offline environments. Pre-purchased UID pools, no network dependency.
Quick integration at low volume: three lines of code and $5 of free credits to start.
Simple, usage-based pricing
One price list across image, audio, and video. Pay only to encode. Verification is always free.
Usage-based
Pay only for what you encode. No minimums, no commitments.
Free, unlimited verification
The EU Code of Practice expects providers to offer detection to deployers, authorities, researchers, and the public. Ours is open to all of them, at no cost.
Volume discounts
Annual commitments for high-volume enterprise clients.
See full pricing
Enter your work email to view per-operation rates, free tier details, and volume discounts.
Or sign in if you already have an account.
Not 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. Three lines of code to start, or an evaluation container for on-prem testing.
Already a customer? Sign in