Chain of Custody for AI Decisions: Cross-Platform Data Provenance, Live in OneConnecter

If you've worked with AI tools in a business context, you already know the compliance problem: data flows through AI, and nobody can explain exactly how.

A marketing manager asks an AI assistant to "pull the top 10 HubSpot contacts from last month and add them to the weekly review sheet". Two minutes later, the data is in Google Sheets. Job done.

Now a regulator asks: "Where did that data come from? Who authorised it? What tools touched it along the way?"

With most AI platforms, the honest answer is: we don't know. The AI made the decisions, the tools executed, and the audit trail — if there is one — shows isolated events with no way to link them together.

That's the gap OneConnecter's chain of custody system closes.

What is a chain of custody for AI?

In legal and compliance contexts, "chain of custody" is the documented trail that proves the integrity of evidence or data from the moment it's collected to the moment it's presented. Every hand-off is recorded. Every handler is known. Nothing is unaccounted for.

OneConnecter applies the same principle to AI workflows. Every step of an AI-driven action — from the initial user query to the final data write — gets tagged with a unique identifier (a chain hash) that links all the steps together. Every tool call, across every platform, inherits that hash automatically. The model doesn't pass it. The user doesn't see it. It just happens.

Chain: 32e5312a-5a49-4240-a596-4fcd482bcb99
  Step 1 [read]   queryData → hubspot_service_agent       — 4,060ms
  Step 2 [source] hubspot-search-objects → 3 contacts     — 1,738ms
Total: 5,798ms

That's a simple read. Chains get more interesting when there are writes, approvals, and multiple platforms involved.

Cross-platform data lineage

Here's where it gets powerful. What if the user reads data from HubSpot, then writes it to Google Sheets? Most audit systems would log two isolated events. OneConnecter links them:

Chain A: 32e5312a... (SOURCE — HubSpot)
  Step 1 [read]   queryData → hubspot_service_agent
  Step 2 [source] hubspot-search-objects → 3 contacts returned

         ↓ data flows cross-platform

Chain B: f7118366... (DESTINATION — Google Sheets)
  source_chain: 32e5312a... ← LINKED TO HUBSPOT
  Step 1 [read]        queryData → google_service_agent
  Step 2 [destination] sheets_append → 3 contacts written

The Google Sheets write row in our audit log literally carries the HubSpot chain hash as its source reference. A regulator clicks the Sheets write → sees source_chain: 32e5312a → clicks through → sees the HubSpot source. Full cross-platform data lineage. One click.

How the governance works

The critical design principle: OneConnecter owns the chain. The model never touches it.

Models are great at reasoning and terrible at bookkeeping. If we relied on the AI to pass chain references through every call, we'd get dropped hashes, hallucinated hashes, and silent compliance gaps. Instead, everything happens server-side:

1. Chain generation — the moment a user query hits OneConnecter, we generate a unique identifier and tag it to the session
2. Automatic inheritance — every subsequent tool call in the same flow inherits that identifier from a server-side cache keyed on the user
3. Cross-instance durability — the cache is backed by a database lookup so it survives serverless cold starts and instance isolation
4. Cross-platform linking — when a user reads from one platform and writes to another within 15 minutes, we automatically stamp the destination write with the source chain

The model just does its job. Compliance happens underneath.

Why this matters: Article 12 with teeth

The EU AI Act's Article 12 requires high-risk AI systems to maintain records of events that allow traceability throughout the system's lifecycle. For any company using AI agents to touch customer data, CRM records, financial data, or regulated communications — that's you.

Most AI platforms answer this with "we log API calls". That's not traceability. That's a list of events.

Traceability means: given any outcome, you can reconstruct the full decision chain that led to it. That's what OneConnecter's chain of custody enables, and as far as we can tell, nobody else is doing it at this level of granularity.

Not HubSpot. Not Salesforce. Not the enterprise AI platforms selling "governance" as a line item.

See it in action

The chain of custody system is live in OneConnecter today — available from the Pro tier upward, with the full cross-platform provenance view on Teams Pro+ and Enterprise.

If you want to see your AI workflows traced the way they should be — start a free account or book a demo. We'll show you a live chain trace of your own data in under five minutes.

---

OneConnecter is an AI orchestration platform for SMEs that need enterprise-grade governance without enterprise-grade cost or complexity. One login, every tool, full audit trail.