AI PROVENANCE INFRASTRUCTURE

Every AI agent is making decisions. Nobody can show their work.

TrustThread.ai gives every AI agent output a structured evidence chain — a Thread — showing exactly what it's based on, what was inferred, what was assumed, and where the trail goes dark.

Built on MCP · x402 · open provenance standards

Provenance graph A chain of nine evidence nodes running left to right. The first three are confirmed, the next two are inferred, the next two are assumed, and the final two are opaque — illustrating where an evidence chain fades from verified fact into assumption. FADE POINT CONFIRMED INFERRED ASSUMED OPAQUE

OPAQUE

The question nobody can answer

An agent finishes a task. It says done. Somewhere downstream, a person, a company, or another agent acts on that claim — approves a filing, books a vendor, ships a decision.

Ask the obvious follow-up: based on what, exactly?

Right now, nobody can answer that in a way that holds up. Not the agent. Not the platform running it. The output looks confident. The logs say it ran. But "it ran" and "it was right" are different claims — and nothing in the current AI stack tells them apart.

CONFIRMED

Meet the Thread

Every AI output gets a Thread — a structured, visual evidence chain you can pull apart node by node. Every claim is classified by how it's grounded: confirmed against something real, inferred through traceable reasoning, openly assumed, or simply opaque because the agent didn't show its work.

One glance answers the only question that matters: how far down does this actually go before it stops being verifiable?

Typed evidence graph

Every node in a Thread is classified — confirmed, inferred, assumed, or opaque. No hidden guesswork, no black-box confidence score.

Human Validation Gate

A named human reviews and signs off, cryptographically, before a high-stakes output ships. Non-repudiable. Append-only. Impossible to fake after the fact.

Built for agents, not just humans

MCP-native. Priced per call. Machine-readable by design — because the next thing asking "can I trust this" is increasingly another agent, not a person.

THE VOCABULARY

We had to invent the words for this

There was no language for AI accountability, so we built one. This is how the next generation of AI infrastructure will talk about trust.

Groundline

The deepest confirmed node in a chain. How far the evidence actually goes before it stops being verified.

Fade Point

The exact node where a chain of solid evidence quietly turns into assumption.

Provenance Debt

The slow accumulation of ungrounded claims an agent racks up over time. Same shape as technical debt.

Shadow Assumption

A hidden assumption buried inside a claim that looks confirmed but isn't.

Human Validation Gate

A formal, named, cryptographically signed human sign-off. Not "the model said it's fine" — a person did, on the record.

Opacity Ratio

The share of a Thread's evidence that the agent simply didn't show. The single most important number nobody currently measures.

WHY NOW

The window is open. It won't stay open.

$146B+

in AI-related M&A over the past year

$281M

raised in AI governance funding in the last 12 months

40%

of enterprise apps will run task-specific agents by end of 2026 (Gartner)

Live

major AI accountability regulation is now in active enforcement

Every other layer of AI infrastructure has already been built — compute, orchestration, evaluation, observability. The provenance layer hasn't. Whoever builds it first owns the vocabulary, the data, and the trust relationships that come with being first.

HOW IT WORKS

Agent runs

Instrument any agent with the Thread SDK or MCP server. Takes minutes, not a sprint.

Evidence chain builds

Every step is classified in real time — confirmed, inferred, assumed, or opaque — as the agent works.

Quality is scored

Depth, confirmed ratio, and opacity ratio, computed automatically the moment the Thread finalizes.

A human signs off

A Validation Gate closes the loop — a named person, on the record, before anything ships.

EARLY ACCESS

Get in before the category has a name

We're opening early access to a small group of builders, operators, and investors who want to see this before it's obvious.

No spam. No spam-adjacent nonsense either. We'll email you when there's something real to see.