When AI goes wrong — and who gets the call
Five cases. One common thread. The accountability gap that courts and regulators are now requiring organizations to close.
The common thread across all five cases: None of these are technology failures. Every one is an accountability failure. An authority failure. A governance failure. Someone built or deployed a capable AI system without establishing, in writing, with named owners, who is responsible for its outputs, who can override it, and who is accountable when it is wrong. That gap is what the courts are finding. That gap is what Colorado and the EU are now requiring organizations to close.
The accountability failures that changed the law
Air Canada argued its chatbot was a separate legal entity responsible for its own statements. The tribunal rejected that defense completely. Air Canada is responsible for all information on its website, including information provided by a chatbot. No named owner for chatbot outputs. No override protocol. No human review layer for consequential customer decisions.
Workday argued it was a software vendor, not an employer. The court rejected that defense. Workday can be held liable as an agent because its customers delegated candidate screening to Workday’s AI. Workday’s own filings disclosed 1.1 billion applications were rejected through its AI tools during the relevant period. Preliminary collective certification granted May 2025 — potentially hundreds of millions of applicants. Complete absence of human oversight. No named person accountable for the screening function.
nH Predict carried an alleged 90% error rate on Medicare Advantage coverage denials — meaning nine out of ten denied claims that were appealed were ultimately reversed. Only 0.2% of patients filed appeals. In March 2026, the court ordered UnitedHealth to disclose the algorithm. The algorithm was allegedly wrong at high rates. The organization allegedly knew. It continued running because the economics worked. No named governance owner with authority to halt the system.
A franchisee operating approximately 110 New York locations filed a $100 million lawsuit alleging a mandated AI delivery optimization system collapsed on-time delivery from above 90% to nearly half of orders taking 45 minutes or more. Alleged sales in New York dropped close to 20 percentage points. Filed May 2026. Mandatory system. No halt authority. No named owner of the system’s performance outcomes. No documented escalation path when outputs began destroying the business.
EU AI Act Article 26 deployer obligations are currently scheduled for August 2, 2026. A pending Digital Omnibus agreement may shift high-risk system obligations to December 2027, but August 2 remains the binding enacted date as of today. Article 50 transparency obligations apply August 2 regardless of the Omnibus outcome. Fannie Mae Lender Letter LL-2026-04, effective August 6, requires a designated AI/ML governance policy owner who reviews the policy at least annually. Colorado SB 26-189, signed May 14, takes effect January 1, 2027. The convergent requirement across all three: a named, accountable owner for each AI system in production, with documented oversight and annual review.
Session notes and complete governance questions — delivered to your inbox
Submit your name and email and the complete five-cases brief comes to you as a PDF. No subscription. No pitch. One document.
- Complete case summaries including rulings and governance failure analysis
- The three questions every board and operating team needs to answer now
- Regulatory deadlines and what each one requires by name
- A pointer to the VTCDO AI Governance Readiness Assessment sample output
- Notification when the next session in the series drops
Five Cases Brief Download
Jay Hawkinson, NACD.DC · VTCDO by Hawksroost · Your information is not shared or sold. You will receive the PDF and occasional session notifications. Unsubscribe at any time.
25 years building data and AI organizations inside industrial manufacturers, CPG companies, and PE-backed enterprises. First functional CDAO at Valmont Industries. Built the 60-person data and AI function at Lamb Weston. Interim CIO at AFL Global. Director of Technology at a PwC operating unit.
The five cases above are what happens at the board and regulatory level when governance fails. The companion VTCDO document maps the same governance gaps against ten specific operating scenarios: quality inspection, demand forecasting, workforce scheduling, predictive maintenance, pricing, supplier assessment, and more.
This document is informational only and does not constitute legal advice. Case characterizations reflect publicly reported facts and allegations; not all matters have been fully adjudicated. Regulatory descriptions reflect the author’s professional assessment as of June 2026. Organizations should consult qualified legal counsel regarding compliance obligations specific to their operations.