Who this is for: GenAI providers, content platform operators, compliance officers, AI product counsel, media and publishing teams. Maps SB 1000 transparency and provenance requirements to cryptographic witness evidence.
Pending Governor Signature
SB 1000 was enrolled and presented to the Governor on September 2, 2026. It is an urgency statute -- it takes effect immediately upon signing. The enrolled bill removes the 1M monthly user threshold, replacing it with coverage of ALL generative AI providers. This crosswalk reflects the enrolled text.
Contents
1. Quick Reference 2. What Does SB 1000 Require? 3. Requirement-to-Procedure Mapping 4. Procedure Detail Cards 5. SB 1000 vs. AB 2013: What Changed 6. Quick Reference: Examiner Questions 7. Quick Start 8. References1. Quick Reference
| Bill | SB 1000 -- California AI Transparency Act |
| Author | Senator Steve Padilla |
| Status | Enrolled September 2, 2026 -- urgency statute (immediate effect upon signing) |
| Scope | All generative AI providers -- no user threshold |
| Predecessor | AB 2013 (California AI Transparency Act, 2024 -- had 1M user threshold) |
| Key change | Replaces "AI detection tool" with "disclosure verification tool" (active provenance, not passive detection) |
| Primary obligation | Latent disclosure embedded in AI-generated content specifying creation vs. alteration |
| SWT3 procedures mapped | 12 |
2. What Does SB 1000 Require?
SB 1000 requires every covered provider to include a latent disclosure in content generated by their system. The latent disclosure must indicate whether the system created or altered the content. This distinction is new -- previous law did not require specifying the type of AI involvement.
The bill replaces the concept of an "AI detection tool" (passive, probabilistic) with a "disclosure verification tool" (active, deterministic). This shift means compliance requires embedded provenance -- think C2PA Content Credentials, not AI classifiers. The verification tool must be able to confirm whether content contains a valid latent disclosure, not guess whether content "looks" AI-generated.
The removal of the user threshold is the most significant change from AB 2013. Previously, only providers with over one million monthly users were covered. The enrolled bill covers ALL providers, including startups, open-source model hosts, and API providers. Any person that creates, codes, or produces a generative AI system -- or substantially modifies one -- is a covered provider.
The latent disclosure must be preserved through "reasonably foreseeable" downstream uses. This creates a chain-of-custody obligation that extends beyond the original generation. Content that loses its provenance data during downstream processing creates a compliance gap for both the original provider and the downstream deployer.
3. Requirement-to-Procedure Mapping
| SB 1000 Requirement | Section | SWT3 Procedure(s) | Evidence Description |
|---|---|---|---|
| Latent disclosure in generated content | Sec. 22758.1 | AI-MARK.1, AI-TRANS.1 | Content provenance marking embedded at generation; transparency disclosure attestation |
| Disclosure verification tool | Sec. 22758.2 | AI-MARK.1, AI-WATERMARK.1 | Provenance verification capability; watermark integrity attestation |
| Created vs. altered distinction | Sec. 22758.1(b) | AI-TRANS.1, AI-EXPL.1 | Transparency disclosure specifies creation/alteration; explanation documents modification scope |
| Provenance data preservation | Sec. 22758.3 | AI-CHAIN.1, AI-PROV.1 | Chain handoff witnessing through downstream uses; model provenance chain |
| Provider identification | Sec. 22758.1(a) | AI-ID.1, AI-TRANS.1 | Agent identity attestation; transparency disclosure includes provider identity |
| System capability disclosure | Sec. 22758.4 | AI-BASE.1, AI-EXPL.1 | Behavioral baseline; explanation generation for system capabilities |
| Content authenticity | Sec. 22758.1 | AI-INF.1, AI-MARK.1 | Inference provenance (prompt/response hash); content provenance marking |
| Downstream use tracking | Sec. 22758.3 | AI-CHAIN.1, AI-LCM.1 | Chain handoff witnessing; lifecycle tracking across downstream consumers |
| User notification | Sec. 22758.5 | AI-TRANS.1, AI-HITL.1 | Transparency disclosure to end users; human review for notification delivery |
| Assistive technology exception | Sec. 22758.6 | AI-EXPL.1 | Explanation generation documents assistive technology classification |
| Compliance documentation | General | AI-AUDIT.1, AI-TRANS.1 | Audit log integrity for compliance records; transparency disclosure chain |
| Enforcement readiness | AG enforcement | AI-AUDIT.1, AI-INCIDENT.1 | Audit trail preservation; incident reporting for AG inquiries |
4. Procedure Detail Cards
Content Provenance Marking
SB 1000 context: The core obligation is latent disclosure -- provenance data embedded in AI-generated content. This replaces passive detection with active, embedded provenance. Think C2PA Content Credentials, not classifiers.
SWT3: Witnesses content provenance marking events. Factor A captures content hash with embedded provenance, Factor B records marking method (C2PA, metadata, steganographic), Factor C flags provenance completeness.
AI-MARK.1 anchors should appear for every content generation event that produces user-facing output. The anchor proves provenance was embedded at generation time, not retroactively added. Check that Factor B indicates a deterministic method -- SB 1000 requires disclosure verification, not probabilistic detection.
Transparency Disclosure
SB 1000 context: Users must be informed that content was AI-generated. The disclosure must specify whether the system created or altered the content. This is separate from the latent (machine-readable) disclosure.
SWT3: Witnesses transparency disclosure events including type, delivery method, and content action (created vs. altered).
Pair AI-TRANS.1 with AI-MARK.1. SB 1000 requires both: latent disclosure (machine-readable, in the content) and notification (human-readable, to the user). A system with AI-MARK.1 but no AI-TRANS.1 satisfies the provenance requirement but may not satisfy the user notification requirement.
Multi-Agent Chain Handoff
SB 1000 context: Provenance data must be preserved through "reasonably foreseeable" downstream uses. When AI-generated content passes through multiple systems (generation, editing, publication), provenance must survive each handoff.
SWT3: Witnesses chain handoff events with shared cycle IDs linking all anchors in a processing chain.
Look for AI-CHAIN.1 anchors at each stage of the content pipeline. The cycle_id should be consistent from generation through publication. Broken chains indicate provenance loss -- a compliance gap under SB 1000's preservation requirement.
Model Provenance Chain
SB 1000 context: The disclosure verification tool must trace content back to its generating system. Model provenance is the foundation -- you cannot verify content origin without knowing which model produced it.
SWT3: Witnesses model provenance from training through deployment, establishing the identity chain for the generating system.
AI-PROV.1 establishes the model identity chain. Combined with AI-INF.1 (inference provenance), this creates a complete trail from model to output. SB 1000's "disclosure verification tool" concept requires this level of traceability.
Agent Identity
SB 1000 context: Provider identification is required in the latent disclosure. The generating system must be identifiable to support the disclosure verification tool.
SWT3: Witnesses agent identity attestation including system identifier, version, and configuration hash.
Every AI-MARK.1 anchor should have a corresponding AI-ID.1 anchor from the same session. Missing identity attestation means provenance cannot be traced to a specific provider -- non-compliant under SB 1000.
5. SB 1000 vs. AB 2013: What Changed
| Provision | AB 2013 (2024) | SB 1000 (2026) |
|---|---|---|
| User threshold | 1M monthly users | ALL providers -- no threshold |
| Detection approach | "AI detection tool" (passive, probabilistic) | "Disclosure verification tool" (active, deterministic) |
| Manifest disclosure | Optional visible watermark/label | Removed |
| Latent disclosure | Required | Required + must specify created vs. altered |
| Assistive tech | No exception | Temporary exception |
6. Quick Reference: Examiner Questions
| Examiner Question | Where to Look |
|---|---|
| How do you embed provenance in AI-generated content? | AI-MARK.1 anchors with Factor B showing embedding method. Must be deterministic, not probabilistic. |
| Can you verify content origin? | AI-PROV.1 model provenance + AI-INF.1 inference hash + AI-ID.1 agent identity create the verification chain. |
| How do you preserve provenance through downstream uses? | AI-CHAIN.1 chain handoff witnessing with consistent cycle_id across pipeline stages. |
| Do you distinguish between created and altered content? | AI-TRANS.1 transparency disclosure with creation/alteration flag in Factor B. |
| What is your disclosure verification tool? | The SWT3 anchor chain IS the disclosure verification tool -- cryptographic proof of provenance, not probabilistic detection. |
| How do you handle the assistive technology exception? | AI-EXPL.1 explanation documents assistive technology classification. Anchors still minted but disclosure obligations are suspended. |
7. Quick Start
w = Witness(
endpoint="https://sovereign.tenova.io/api/v1/witness",
api_key="axm_live_...",
agent_id="content-generation-pipeline"
)
# Witness content provenance marking (SB 1000 latent disclosure)
anchor = w.record(
procedure="AI-MARK.1",
factor_a="content_sha256:7f83b165...",
factor_b="method:c2pa_embedded",
factor_c="provenance_complete"
)
print(f"Content provenance: {anchor}")
# Witness transparency disclosure (SB 1000 user notification)
disclosure = w.record(
procedure="AI-TRANS.1",
factor_a="disclosure_type:ai_generated",
factor_b="content_action:created",
factor_c="notification_delivered"
)
print(f"Transparency disclosure: {disclosure}")
Full SDK documentation: sovereign.tenova.io/docs
Create a free account: sovereign.tenova.io/signup
8. References
- California SB 1000 Enrolled Text (Legislature.ca.gov)
- SB 1000 Bill Status
- C2PA Content Credentials Specification
- Related: California AB 2013 (Training Data Transparency)
- Related: California SB 942 (AI Transparency Act)
- Related: California SB 53 (Frontier AI Transparency)
- Related: EU AI Act Article 50 Transparency
- SWT3 UCT Registry, 280 procedures across 9 namespaces
- SDK Documentation