The Mark Didn’t Survive

“The vendor added the mark. Is that our AI inventory — or just our vendor list?”

The question this session is built around.

The question is no longer whether the vendor marked the output. It is whether your organization can show that the mark survived until the content reached the public — and who is accountable if it didn’t.

Inventory one — usually exists

Systems

What AI systems exist, when each was placed into service, who supplies it, and who deploys it. This is the list procurement maintains, and it is where most AI governance programs stop.

Inventory two — rarely exists

Content workflows

Which systems produce text, images, audio, or video. Which of those outputs are published externally. What happens to the output between generation and publication — and at which of those steps a machine-readable mark can be stripped.

Inventory three — almost never exists

Capabilities

Not which tools you approved — what those tools can actually do. Whether a capability is disabled technically or only prohibited by policy. Whether any administrator, employee, or third party can reach it. And who has the authority to switch it off across the enterprise before the end of the day.

A capability that arrived inside an approved platform never got its own inventory line. That is the gap the session closes.

Everyone votes before anyone senior speaks

Each decision-maker commits privately before discussion opens. The sponsor votes last, by rule. The gap between the first vote and the last is the finding.

The functions answer only when asked

Vendor Risk knows what the attestation covers. Information Security knows what testing was done. Content Operations knows which publishing route each business unit actually uses. None of them volunteer it.

One round ends in exact wording

The room drafts the publication rule word for word: which systems are covered, which workflows are permitted, what visible label is required, who verifies marking, who may grant an exception, and who holds the evidence.

The final round raises the stakes

Late in the hour the room learns that a legacy module inside an approved platform includes a capability falling within a newly prohibited content category — a capability that never appeared in the inventory, because the platform did. The decision is no longer about labels. It is about who can switch something off, how fast, and on what evidence.

This is an operational exercise in the workflow your organization owns. It is not a legal briefing, the facilitator reaches no legal conclusions, and nobody is scored.

A — Vendor-based review

Request transparency attestations from all approved vendors and proceed on that basis. The vendors built the systems and are best placed to evidence their controls.

B — System-and-use review

Classify each system and identify which business units use it for what. Attestations mean little without knowing where and how each system is actually used.

C — End-to-end content inventory

Map systems, outputs, transformations, publication channels, and who owns the final disclosure. The obligation attaches to what reaches the public, not to what the vendor generated.

All three are defensible in a real room, and each costs a different amount of time the organization may not have. Which one your team picks — before and after consulting the functions three feet away — is the data.

The room does not yet know that the mark does not always survive the workflow.

A content-marking control chain — generation, transformation, publication, verification, and retention, with an owner at each step.

An inventory ownership map — who owns systems, who owns content workflows, and who owns capabilities, which is usually the first time those three have been separated.

A vendor-attestation standard — what your organization accepts, what it independently verifies, and what it treats as incomplete.

A capability-response standard — who can restrict access to a feature enterprise-wide, what evidence is required, and who can restore it.

Initial-versus-final vote data for every round, and a consultation log showing which functions were asked, when, and what.

A readiness register with named owners and dates — and everything classified conservatively as exercise observations and validation questions, never “findings.” You review the draft summary before anything is final.

Where this sits

The Mark Didn’t Survive is part of the Xcelus Decision Brief™ — Leadership Edition, a set of 60-minute facilitated exercises each built for a different leadership table.

Its companion is August 2 Didn’t Move, which tests whether leaders can separate deferred requirements from those already in force. That session asks what could stop. This one asks who owns what you shipped. See all current sessions →