The AI Owner Isn't on the Org Chart

Show notes

IBM asked 2,000 technology executives about the AI running in their companies. Two-thirds said the same thing: I'm accountable for AI systems I don't fully control. That's the quiet crisis sitting inside every "AI operating model" slide — a Chief AI Officer wave (26% → 76% of companies in one year), a council, a RACI with someone's name in the A column, and underneath it all, systems arriving weekly through business units and personal accounts that no owner can see, stop, or budget for. This week: why naming an owner isn't the same as having one — and what the operating model looks like when someone actually has to sign.

This episode of Above the Noise — the unbiased AI brief for enterprise leaders:

• News Brief: The EU AI Act's August 2 deadline arrived, and most teams read the wrong headline — high-risk obligations slipped to Dec 2027/Aug 2028, but chatbot disclosure, AI-content marking, deepfake labeling, and the Commission's GPAI enforcement powers all landed on schedule, with fines up to €15M or 3% of global revenue. Enforcement starts where proof is cheap · Akamai's new browser-telemetry research: 47% of enterprise AI conversations run through personal accounts — whatever your AI inventory says, the real number is roughly double · Google restructured its own AI ownership — DeepMind's CEO title is gone and the operator now reports straight to Pichai. When the bet gets big enough, accountability moves up, not out.
• Expose a Lie ⭐: "We named someone to own AI, so AI is owned." Three breaks: the name doesn't come with control (66% of tech CxOs accountable for AI they don't fully control; 70% say business deploys faster than IT can track; 70% of CEOs claim they own AI strategy while 6% are in the decisions) · the name doesn't come with eyes (half the estate invisible, 85% can't see AI spend in real time, an average of 54 AI agent incidents per organization last year) · and the name is parked at the wrong altitude — incidents don't happen to "AI," they happen to decisions, and the council owns the strategy while Tuesday 2 a.m. goes unowned. Accountability assigned without control doesn't disappear. It waits — that's blame, pre-positioned.
• The Playbook: The operating model when someone has to sign. A real owner is a name plus sight, say, and budget — and IBM's data says embedded control is a speed play, not a tax: 25% fewer incidents, 16x more agents deployed, 18% higher margins, at a quarter of the AI budget spend. Three moves: count before you assign (the register — every system, four fields, including "who can turn it off") · assign decisions, not systems (the three-bucket test: acts alone / asks first / never touches) · arm the owner (real-time visibility, launch veto, working kill switch, and a controls budget of their own). The Line for the Meeting: "Who can stop the launch? That's the owner. Everyone else is advice."
• The question to sit with: Your most consequential AI system misfires tonight. Whose phone actually rings — and did that person agree to own it before tonight, with sight, say, and budget? If the owner finds out they're the owner during the incident, you never had an owner. You had a name in a blast radius.

No vendors. No hype. Just the signal.

👉 Follow Above the Noise wherever you listen — new episode every two weeks. Send it to one leader whose name is in the A column of a RACI they've never seen, and leave a rating so more people find it. Between episodes, find Shaun on LinkedIn — come argue with him there. Especially if your Chief AI Officer has no budget and no kill switch. He wants to hear how that's going.