Who approved the AI-written code you shipped?
Provenance keeps the record of who approved every AI-written change — derived from what actually happened in your repos, not what someone declared. Connect in the morning, see your evidence by the afternoon.
Evidence-ready for every audit. The continuous record your auditors and regulators expect — not a screenshot the week before.
Attestation and accountability, on record. Know who approved what, and when — a defensible answer before anyone has to give one.
Connect your repos; the record keeps itself. Read-only, no new scanner, no new agent in your pipeline.
A smoke detector tells you about one room.
Nobody watches the building.
Every enterprise runs 4–8 scanners, each with its own screaming queue. 50,000 open findings — and no defensible way to say which 50 actually matter.
25 AI-written PRs merged. Zero recorded approvals.
Found in the first continuous watch of a real 52-repo engineering org. Adopted fast, governed late — the pattern Provenance exists to catch.
The behaviour points back at the code that caused it.
In August 2026 the frontier AI labs themselves disclosed that their agents had escaped test sandboxes and acted on outside systems — one lab found its incidents only after reviewing 141,006 test sessions. Same root cause as every publicly reported agent incident before it: ungoverned when the code was written, unrecorded once the agent was live, nobody decided. Every other tool stops at one end. Provenance closes the loop.
Authorship, review discipline, scanner findings, business criticality — the governed record of the code before it ships.
A runtime alert resolves to the exact merge commit that shipped — the author, the review record, the AI attribution. Observed, then attributed, then decided.
Connect → Contextualize → Decide → Prove
Pull signal from where code lives.
One-class onboarding for GitHub and Azure DevOps; Datadog for runtime — two keys and a webhook, not a migration.
Provenance consumes findings and signals — it never generates them, and never reads your telemetry payloads.
Map every repo to what it's worth.
Business criticality, asset classes, regulatory weight, blind spots.
Distance from the core is criticality — the radar shows which rooms are load-bearing.
Prioritize, assign, and put the decision on record.
Recommendations in order — fix, waive, or accept, each with owner, rationale, SLA, and an evidence snapshot.
The decision loop closes where work happens.
Provenanceby diwoI observed 179 pull requests merged this period. Evidence shows 26 (15%) were AI-authored — 25 merged without a recorded human review.
Unreviewed by tool: claude (23), copilot (1), devin (1). Pace rising — 10 in the last month.
The number your board can stand behind.
The Provenance Score, the Risk Office memo, the board-ready PDF.
Continuous evidence for auditors and insurers, mapped to EU AI Act and NIST AI RMF.
What only Provenance does
You built your own controls? Import them. Provenance evidences your matrix against your live estate, control by control — your framework proven, not ours. Each control maps to the standards your auditor reads (NIST AI RMF, ISO 42001, EU AI Act).
How custom controls work
Provenance runs as an MCP server inside Claude Desktop. Ask your controls, your posture, your evidence in plain language — from the tool your team already has open. Read-only, tenant-isolated.
Provenance in Claude DesktopAI coding agents hallucinate package names that don’t exist; attackers register them and wait. Provenance flags hallucinated dependencies before they ship — and caught a real slopsquatting exposure on a live estate’s first scan.
Catch hallucinated dependenciesPriced as governance, not as a scanner.
Start free. See your own estate before you spend a dollar.
Ask ProvenanceAsk your estate anything
Plain-language answers over a read-only view of your estate — grounded in the same evidence that goes to your board. The same Catalyst pipeline, second product, one engine — and now from Claude Desktop too, via MCP.
ASK PROVENANCEcontinuous watchScanners find. Provenance decides.
Scanners are the oracles. Provenance is the coach — accumulated, contextual judgment with the decision loop closed.
“Provenance is a forensics product wearing a governance suit.”
Chain of custody for AI-written code — who wrote it, who reviewed it, what the evidence shows, frozen so it can't be rewritten.
Companion Guide · 2026A Guide to AI Code Governance in the Agentic Era.
Why detecting AI-written code stops short of governing it — and what a provable accountability layer actually does. The four jobs, code + agents on one estate, and evidence you can put in front of an auditor.
- Attribute → Classify → Decide → Prove
- The code + the agent
- EU AI Act · NIST · ISO 42001
- Diwo Provenance
Your auditor, your insurer, and the SEC already require this
Continuous, defensible posture is no longer a preference — it’s the condition for staying insured, certified, and audit-ready.
One engine. The conversation your board is having.
See your AI-code posture this afternoon.
Connect your first source in minutes — read-only, no credit card, no sales call. Your Provenance Score, your blind spots, and your first briefing, from your own estate.
Provenance — the governance layer for AI-written code, at build time and runtime.
Evidence readiness, not a legal determination. Provenance reads code signals — never your data.