Executive Research · AI Governance · May 2026

When AI fails publicly,

the brand pays the bill.

Chander Dhall Builder • Leader • Speaker

From Air Canada's chatbot inventing refund policies to deepfake video calls authorizing $25M transfers, generative AI failures are no longer hypothetical. They are documented, litigated, and expensive. The question is not whether your AI will make mistakes. It is whether your governance can catch them before they reach the customer.

Read Full Report → Start a Conversation

7Failure mechanisms

Executive Snapshot

Four modalities. One pattern. Zero tolerance.

Text, audio, image, and video AI systems have all failed publicly in the last three years. Each failure shares the same root cause: the system was deployed without the governance controls that would have caught the error before it reached the customer.

Text / Chat Hallucination

Air Canada chatbot invented a refund policy. Tribunal held the airline liable.
2024

Audio / Voice Deepfake

AI voice impersonating Biden used in robocalls. FCC enforcement followed.
2024

Image Inaccuracy

Google paused Gemini image generation after historically inaccurate outputs.
2024

Video $25M

Deepfake video call impersonated executives. Finance worker authorized transfer.
2024

Text / Chat Failures

The chatbot said it. Now you own it.

Customer-facing chatbots have hallucinated refund policies, sworn at customers, advised businesses to break the law, and provided harmful health guidance. In each case, the organization was held responsible for the output.

Air Canada

Liable

Chatbot invented refund policy

Customer relied on chatbot response about bereavement fares. Tribunal held airline responsible for misinformation.

DPD

Viral

Chatbot swore at customers

Customer prompted support chatbot into profanity and insults. DPD disabled it after incident went viral.

NYC MyCity

Illegal

Chatbot advised breaking law

NYC business chatbot produced incorrect guidance including advice that could violate regulations.

LLM as Expert Failures

Looks right is not evidence.

When LLMs are used for legal research, compliance advice, or decision support, the failure mode shifts from embarrassment to liability. Fabricated citations, leaked confidential data, and insecure code have all reached production.

Legal Filing

Sanctioned

Lawyers filed fake citations

Legal filing included fabricated case citations and quotes from ChatGPT. Court sanctioned the lawyers involved.

Data Leakage

Banned

Employees pasted secrets

Reports of employees pasting confidential code and data into ChatGPT triggered internal bans at Samsung and others.

Code Security

Worse

AI code less secure

Studies report AI assistance can increase vulnerability rates and reduce secure-by-default behavior, especially for novices.

Audio / Video Failures

I saw them on Zoom is no longer a control.

Voice cloning and deepfake video have moved from research curiosity to operational threat. McDonald's ended a voice AI pilot after ordering errors. Deepfake video calls have authorized millions in fraudulent transfers.

Voice Ordering

Ended

McDonald's AI drive-thru pilot

Voice AI test ended after inconsistent ordering performance. Speech-to-intent fails at accents, noise, and ambiguity.

Voice Deepfake

Illegal

Biden robocall impersonation

AI-generated voice impersonating President Biden used in robocalls. FCC enforcement actions followed.

Video Deepfake

$25M

Deepfake video call fraud

Fraudsters used deepfake video to impersonate executives in calls, leading to unauthorized transfers.

Image Failures

Synthetic images create expectation debt.

AI-generated images have caused brand damage when historically inaccurate outputs went viral, and operational chaos when marketing assets promised experiences that could not be delivered.

Google GeminiImage generation paused

What happenedFebruary 2024

Paused

Root causeHistorically inaccurate outputs

Brand

ImpactViral criticism, feature disabled

Trust

Willy's Chocolate ExperienceAI marketing disaster

What happenedFebruary 2024

Viral

Root causeAI images vs. reality mismatch

Fraud

ImpactRefunds, police called, global news

Crisis

Failure Mechanisms

Seven mechanisms. Every modality. Same patterns.

These failure mechanisms are portable across text, audio, image, and video. If your governance does not address each one, the failure is a matter of when, not if.

Hallucination as policyConfidently wrong answers become product promises

Legal

Prompt injectionOutputs and actions bypass intent and policy

Security

Ambiguity at the edgesAccents, sarcasm, mixed intents, partial context

Ops

Over-automationNo human fallback means small errors become incidents

Scale

Weak provenanceNo citations, no source-of-truth linkage

Trust

Data leakageEmployees paste secrets; logs retain sensitive data

Compliance

Model driftVendor updates cause silent behavior changes

Contract

Governance Controls

Eight controls. One governance posture.

Each control maps to one or more failure mechanisms. Together, they form the minimum viable governance for any production AI system.

Control	What it means	Failure it prevents
System-of-record rule	Model never invents policy; only quotes approved sources	Hallucination as policy
Evidence-first UX	Show citations, snippets, last-updated date; allow "open source"	Weak provenance
Tiered automation	Low-risk automated; high-risk requires human review	Over-automation
Eval gates	Offline eval sets, adversarial tests, red-teaming, phased rollout	Ambiguity at edges
Conversation firebreaks	Detect jailbreaking, isolate tools, enforce allow-lists	Prompt injection
Human fallback	Explicit escalation path, SLAs, ownership, incident response	Over-automation
Audit logging	Who/what/when, prompts, sources, actions, retention policy	Data leakage
Change management	Model/version pinning, regression tests, release notes	Model drift

Board Readiness

Two questions separate governance theater from production control.

These questions expose the gap between vendor capability claims and what actually happens when the system reaches a customer.

Question 01 Source of truth

Can your AI system cite the approved source for every claim it makes to a customer?

Why it matters Liability

If the system can invent policy, you own the policy it invents. Air Canada learned this in court.

Question 02 Human fallback

When the AI fails, how fast can a human take over, and who owns that handoff?

Why it matters Blast radius

Without explicit escalation, small errors become incidents at the speed of automation.

Call to Action

The Chander Dhall AI Governance Review.

A focused engagement to map your AI systems against the eight governance controls, identify gaps, and build a remediation roadmap before the next failure becomes yours. Built for CEO/CFO/CIO and board review.

Start a Conversation →

1

Inventory

Map every AI system touching customers, employees, or sensitive data. Identify which governance controls are present, partial, or missing.

2

Gap analysis

Score each system against the eight controls. Prioritize by blast radius: customer-facing, high-value decisions, and regulatory exposure.

3

Remediation roadmap

Phased plan to close gaps. Technical controls, process changes, and vendor requirements. The deliverable is production readiness, not a deck.

The Governance Question

Every AI failure in this deck
was preventable.

The organizations that avoid the next Air Canada, the next $25M deepfake, the next viral chatbot disaster, are the ones that ask the governance questions before they deploy.

Read Full Report