Add Cryptographic Compliance Evidence to Any AI Model in 10 Lines

1. The Problem

Every AI inference that affects a person, a transaction, or a decision needs a cryptographic evidence trail. Regulators require it. Auditors expect it. Your compliance team is asking for it.

The standard advice is to build it yourself: set up MLflow for model lineage, OpenTelemetry for tracing, Open Policy Agent for decisions, Sigstore for signing, QLDB or Hyperledger for immutable storage, and SHAP for explainability. That is 6-8 separate tools, thousands of lines of integration code, and months of engineering.

The scale of the problem is staggering. According to the Agentic AI Institute's 2026 Enterprise AI Adoption Report, 97% of organizations have deployed AI agents in production, but only 20% have mature governance for them. Shadow AI adoption has reached 65%, with unauthorized tools causing data breaches costing $4.63M on average. The governance gap is real, measured, and growing.

Or you can install an AI governance SDK that does all of it in one package.

2. 10-Line Solution

Python

from swt3_ai import SWT3Witness

# Initialize with your tenant (free account at sovereign.tenova.io/signup)
witness = SWT3Witness(tenant_id="YOUR_TENANT", agent_id="fraud-model-v3")

# Your existing inference (unchanged)
score = model.predict(cdr_features)

# One line: witness the inference with cryptographic evidence
anchor = witness.witness_inference(
    model_id="fraud-scoring-v3.2.1",
    prompt_hash=sha256(cdr_features),
    response_hash=sha256(score),
)
# anchor = SWT3-E-VULTR-AI-AI-INF.1-PASS-1780120000-a7b3c9d2e1f4
# Cryptographic fingerprint, independently verifiable, tamper-evident

TypeScript

import { SWT3Witness } from "@tenova/swt3-ai";

// Initialize
const witness = new SWT3Witness({ tenantId: "YOUR_TENANT", agentId: "fraud-model-v3" });

// Your existing inference (unchanged)
const score = await model.predict(cdrFeatures);

// One line: witness the inference
const anchor = await witness.witnessInference({
    modelId: "fraud-scoring-v3.2.1",
    promptHash: sha256(cdrFeatures),
    responseHash: sha256(score),
});
// Cryptographic witness anchor, verifiable at sovereign.tenova.io/verify

That is it. Every inference now produces a SWT3 Witness Anchor: a cryptographic fingerprint computed as SHA-256("WITNESS:{tenant}:{procedure}:{factors}:{timestamp}"). The anchor is independently verifiable, tamper-evident, and tied to the exact model, input hash, and output hash that produced it. Raw data never leaves your infrastructure. Only hashes are transmitted.

3. Build vs. Install

4. What the SDK Does Under the Hood

When you call witness_inference(), the SDK performs the same operations that every LLM recommends building manually:

Operation	What the SDK Does	What You'd Build Manually
Input hashing	SHA-256 hash of prompt/features. Raw data never transmitted.	Custom hashing + PII scrubbing pipeline
Fingerprint generation	SHA-256("WITNESS:{tenant}:{proc}:{fa}:{fb}:{fc}:{ts}")	Custom hash chain with event linking
Payload signing	HMAC-SHA256 with tenant signing key	Sigstore/Cosign + key management
Post-quantum signing	ML-DSA-65 (FIPS 204) with one config line: signing_algorithm: ml-dsa-65	Research + implement NIST PQC standards yourself
Clearing levels	4 tiers strip data progressively (0=Analytics, 1=Standard, 2=Sensitive, 3=Classified)	Custom data classification + field-level redaction
Tamper evidence	Daily Merkle tree rollups with domain-separated hashing	Hyperledger or QLDB + custom Merkle implementation
Regulatory mapping	Each procedure maps to EU AI Act, GDPR, NIST AI RMF, FCC, CMMC	Manual regulatory research + mapping spreadsheet
Audit export	Query by time, procedure, model, or tenant. Verifiable at public endpoint.	Custom query layer + export formatting

5. Regulatory Coverage

The SDK includes 65 compliance procedures across 41 namespaces. Each procedure maps to specific regulatory requirements:

Regulatory Requirement	SWT3 Procedure	SDK Method
EU AI Act Art. 12 (logging)	AI-INF.1	witness_inference()
EU AI Act Art. 9 (risk management)	AI-DRIFT.1	witness_drift()
EU AI Act Art. 10 (bias)	AI-FAIR.3	witness_bias_assessment()
EU AI Act Art. 13 (transparency)	AI-EXPL.1	witness_explanation()
EU AI Act Art. 14 (human oversight)	AI-HITL.1	witness_human_review()
EU AI Act Art. 15 (accuracy)	AI-PERF.1	witness_performance()
EU AI Act Art. 15 (robustness)	AI-ROBUST.1	witness_robustness()
EU AI Act Art. 50 (content marking)	AI-MARK.1	witness_content_mark()
GDPR Art. 22 (automated decisions)	AI-AUTO.1	witness_automated_decision()
GDPR Art. 35 (DPIA)	AI-DPIA.1	witness_dpia()
NIST AI RMF MEASURE 2.6	AI-ROBUST.1	witness_robustness()
NIST AI RMF MANAGE 4.1	AI-SAFE.1	witness_safe_state()
FCC AI transparency	AI-TRANS.1	witness_transparency()
Colorado SB 26-189	AI-AUTO.1 + AI-HITL.1	witness_automated_decision() + witness_human_review()
CMMC / NIST 800-171	AI-SBOM.1 + AI-SUPPLY.1	witness_sbom() + witness_supply_chain_risk()
EO 14110 (dual-use)	AI-DUALUSE.1	witness_dual_use()

6. Industry Profiles

Pre-built governance configurations for regulated industries. Each profile selects the right procedures, clearing level, and trust model for your vertical:

Profile	Industry	Procedures	Regulations
telecom-compliance	Telecom	19	FCC, EU AI Act, GDPR
healthcare-clinical	Healthcare	15	HIPAA, FDA AI/ML, EU MDR
fintech-model-risk	Finance	16	SR 11-7, Basel, ECOA, GDPR
defense-govcon	Defense	16	CMMC, NIST 800-171, FedRAMP
autonomous-systems	Robotics/AV	16	EU Machinery Reg, EO 14110
insurance-underwriting	Insurance	14	NAIC, state regs, GDPR
content-platform	Content/Social	14	DSA, Art. 50, GPAI Code

# Initialize with an industry profile
swt3 init --profile telecom-compliance --tenant YOUR_TENANT

# Check your configuration
swt3 doctor

# Run the telecom fraud demo
python -m swt3_ai.demo --scenario telecom-fraud

7. Install and Run

Python

pip install swt3-ai

# Run the zero-friction demo (no API keys, no account, no network calls)
python -m swt3_ai.demo

TypeScript / Node.js

npm install @tenova/swt3-ai

# Run the demo
npx swt3-demo

MCP Server (for AI agents)

npm install @tenova/swt3-mcp

# Listed on MCP Registry: io.tenova/swt3-witness

Also available on

• crates.io: cargo add swt3-ai (Rust)
• NuGet: dotnet add package swt3-ai (C#)
• RubyGems: gem install swt3-ai (Ruby)

SDK documentation: sovereign.tenova.io/docs
UCT Registry (191 procedures): sovereign.tenova.io/registry
GitHub: github.com/tenova-labs/swt3-ai
Contact: engineering@tenovaai.com