AI Governance
Govern AI decisions with the rigor they demand
Most AI governance tools stop at the model. DecisionLedger governs the decision the AI informs. Inventory every model and agent, run risk and impact assessments, enforce human-in-the-loop policies at runtime, and keep the immutable audit trails regulators require, with controls mapped to the EU AI Act, NIST AI RMF, and ISO/IEC 42001.
Challenges We Solve
Governance gaps that structured controls and audit trails eliminate.
Shadow AI Sprawl
Unsanctioned AI tools proliferate across teams with no inventory or risk scoring. Shadow AI Discovery inventories every tool and scores risk per team, while Agent Authority Boundary ensures agents stay inside their delegated envelope.
Prompt Injection & Jailbreaks
Deployed AI surfaces are vulnerable to adversarial inputs that bypass safety controls. Prompt Injection Exposure quantifies your attack surface, and LLM Jailbreak Resilience stress-tests defenses against evolving jailbreak corpora.
Missing Accountability & Provenance
When AI produces a bad output, tracing it back to the responsible prompt, model version, and human edit is impossible. AI Output Attribution Chain and Model Provenance Attestation close this gap.
Rubber-Stamp Oversight
Human-in-the-loop checkpoints exist on paper but don't change outcomes in practice. Human-in-the-Loop Efficacy measures whether HITL reviews actually alter decisions or just add latency.
Models You Did Not Build Here
Your own models run on Databricks, SageMaker, Vertex or a laptop, and your governance record stops at the platform boundary. Register them as first-class governed entities that decisions and lineage can cite, without DecisionLedger executing or evaluating any of them.
AI Governance Pipeline
Every AI agent follows a four-stage governance lifecycle. Each stage is logged, scored, and auditable.
Register
Discover and inventory every AI agent, model, and tool across the organization with automated risk scoring.
Monitor
Continuously track agent behavior, output quality, and policy compliance in real time.
Govern
Enforce authority boundaries, approval gates, and human-in-the-loop checkpoints before actions execute.
Audit
Immutable logs capture every decision, input, output, and override for regulatory and fiduciary review.
Use Cases
Real governance scenarios powered by DecisionLedger.
Runs Shadow AI Discovery across all teams, uncovering 23 unsanctioned AI tools. Scores each with Agent Authority Boundary, then uses Autonomy Graduation Readiness to determine which agents are safe for autonomous operation.
Full AI inventory with risk tiers - 8 tools flagged for remediation, 4 graduated to autonomous
Uses AI Evaluator Calibration to detect a 12% drift in their LLM-as-judge scoring. Runs Human-in-the-Loop Efficacy to prove HITL reviews are changing outcomes. Generates EU AI Act evidence from Model Provenance Attestation and AI Output Attribution Chain.
Calibration drift caught before regulatory review - compliance evidence generated automatically
After a prompt injection incident, runs Prompt Injection Exposure across all AI surfaces and LLM Jailbreak Resilience stress tests. Uses Tool Call Chain Risk to identify agent workflows with unsafe rollback paths, then deploys the kill switch on high-risk agents.
Attack surface quantified, 3 vulnerable surfaces hardened, incident contained with full audit trail
Reviews the platform's exposure to the bank's own credit models before signing. Reads the exact non-liability text a signer affirms, confirms the evaluation boundary is enforced by a code guard rather than a contract clause, and signs the third-party variant for the two vendor models the bank does not own.
Customer-owned models governed in the decision record with no implied warranty by the platform
Imports four Hugging Face models the team already uses, each pinned to a commit SHA rather than a mutable main branch, then links them to the decisions they informed so an auditor can trace which exact bytes were behind each one.
Model references that still resolve to the same weights a year later
Measurable Impact
Based on platform benchmarks across early adopters.
Shadow AI Inventory
Prompt Injection
HITL Effectiveness
Model Provenance
Agent Inventory
Bias Detection
Your Own Models
Hub Model References
Built-in AI Governance Controls
Not just models about AI risk - actual enforcement tools that register, monitor, and control every AI agent and model in your organization.
Agent Registry
Register every AI agent with per-agent permissions, activity monitoring, and instant suspension. Know exactly what AI is doing across your organization.
Kill Switch
Circuit breaker for any AI model or agent. One-click disable across your entire tenant with automatic cool-down re-enable when you're ready.
Shadow Mode
Test new models in production without affecting live decisions or audit logs. Validate AI outputs side-by-side before going live.
Bias Auditing
Statistical bias detection across protected classes with a dedicated bias dashboard. Surface disparate impact before it becomes a compliance finding.
SHAP Explainability
Every prediction comes with clear feature-importance breakdowns showing how inputs drive outputs. No more black-box AI decisions.
Policy Guardrails
Define enforceable constraints in plain rules. Auto-flag, block, or escalate violations with full override tracking and rationale logging.
Drift Detection
Monitor policy and guardrail effectiveness over time with automated alerts when controls go stale. Detect when regulatory or policy changes invalidate existing controls.
Compliance Badges
Pre-mapped controls for EU AI Act, NYC Local Law 144, SOX 302/404, and DOL Fiduciary Rule. Generate compliance evidence automatically from your governance activity.
External Model Registry
Catalog a model you own and operate elsewhere, on Databricks, SageMaker, Vertex, Azure ML, Hugging Face or in-house, as a governed entity with its own version, provenance and lifecycle. Decisions cite it, lineage traces it, and archiving is blocked while anything still depends on it.
A Boundary Enforced In Code
DecisionLedger does not evaluate a registered model for bias, drift, sensitivity or accuracy, and does not execute it. That is not a disclaimer in the contract: a single guard keeps a registered model's identity out of every evaluator and scheduler, and skips with a log if one ever collides.
Signed Non-Liability Attestation
A model cannot go active until an authorized signer e-signs the attestation under ESIGN and UETA. The signed PDF is content-hashed and stored immutably as a legal record with the signer's name, title, timestamp, IP and user agent, in two variants depending on whether the model is yours or a third party's.
Hugging Face Import, SHA Pinned
Search the Hub from inside the platform and import a model without retyping its metadata. A Hub repo has no versions and its main branch can be force-pushed, so every import is pinned to a commit SHA. No weights are downloaded and the browser never talks to huggingface.co.
Connects With
Part of 150+ native integrations across CRM, marketing, finance, HR, ecommerce, and analytics
Salesforce
HubSpot
Stripe
Shopify
Google Analytics 4
Workday
QuickBooks
Snowflake
Slack
Zendesk
GitHub
Meta Ads
Mailchimp
NetSuite
Jira
Power BISalesforceHubSpotStripeShopifyGoogle Analytics 4WorkdayQuickBooksSnowflakeSlackZendeskGitHubMeta AdsMailchimpNetSuiteJiraPower BIFeatured Models
Pre-built decision models ready to run with your data.
Algorithmic Impact Assessment
Formal AIA as quantitative decision model with statistical bias analysis, EU AI Act article scoring, and remediation roadmap
Agent Authority Boundary
Scores whether an AI agent's proposed action falls inside its delegated authority envelope.
Prompt Injection Exposure
Quantifies organizational exposure to prompt injection across deployed AI surfaces.
AI Evaluator Calibration
Detects drift in LLM-as-judge scoring relative to human ground truth.
Model Provenance Attestation
Verifies model lineage, training data attestation, and supply-chain trust.
Agentic Workflow Replay Risk
Scores reproducibility risk of agent decisions when re-executed against changed external state.
RAG Grounding Quality
Measures retrieval grounding strength, hallucination probability, and citation faithfulness.
AI Cost Attribution Anomaly
Detects anomalous per-team or per-feature LLM spend.
Shadow AI Discovery
Inventories unsanctioned AI tool usage with risk scoring per tool/team.
Human-in-the-Loop Efficacy
Measures whether HITL checkpoints actually change outcomes versus rubber-stamping.
Autonomy Graduation Readiness
Decides whether an AI workflow has earned the right to move from supervised to autonomous operation.
Multi-Agent Check Bypass Detection
Detects when multiple AI agents converge on outcomes that bypass intended controls.
Tool Call Chain Risk
Scores risk in agent tool-call chains based on depth, side-effect surface, and rollback feasibility.
AI Output Attribution Chain
Tracks which prompt, model version, retrieval source, and human edits produced a given output.
Synthetic Data Provenance
Detects synthetic content used as training input or evidence with provenance tracking.
Appeal Process Documentation
Captures the customer's appeal channel, response SLA, and review record.
Fine-Tuning Data Leakage Risk
Scores extraction risk of confidential or PII training data from fine-tuned models.
Agent Termination Safety
Pre-deployment scoring of whether an agent can be safely halted mid-execution.
LLM Jailbreak Resilience
Tests deployed AI surfaces against jailbreak corpus and tracks resilience drift.
AI Vendor Lock-In Exposure
Scores switching cost across model providers including prompt portability and contract terms.
AI Incident Postmortem Completeness
Scores AI incident postmortems against structured template.
How It Works
Three steps to structured, auditable decisions.
Discover & Score Risk
Run Shadow AI Discovery to inventory unsanctioned tools, then score each agent with Agent Authority Boundary and Prompt Injection Exposure to quantify your attack surface and authority gaps.
Validate & Harden
Stress-test with LLM Jailbreak Resilience, verify RAG Grounding Quality for hallucination risk, and audit Tool Call Chain Risk to ensure agent workflows can be safely rolled back.
Monitor & Govern
Track AI Evaluator Calibration for scoring drift, measure Human-in-the-Loop Efficacy to prevent rubber-stamping, and detect Multi-Agent Check Bypass before controls are circumvented.
Graduate & Audit
Use Autonomy Graduation Readiness to decide when workflows can move from supervised to autonomous. Maintain full provenance with AI Output Attribution Chain and Model Provenance Attestation.
How many AI agents in your organization have been tested against prompt injection, have verified provenance, or have measured whether human oversight actually changes outcomes? And of the models you built yourself, how many appear anywhere in that record?
Spreadsheet AI inventories
Static model lists with no risk scoring, authority boundaries, or jailbreak testing
Manual compliance documentation
No automated provenance attestation or output attribution chain
MLOps tools without governance
Track model versions but not prompt injection exposure, HITL efficacy, or agent authority
GRC add-on modules
Generic risk tools that don't understand agent tool chains, RAG grounding, or multi-agent bypass risks
Vendors who offer to score your model
An evaluation you did not ask for, on a model they did not build, creating a warranty nobody wanted