Your AI Policy Describes a Company That Doesn't Exist

The conversation delves into AI governance in regulated industries, the gap between what governance policies say and what employees actually do, the human-in-the-loop concept as it works in practice rather than on a process map, and where accountability sits when an AI writes something and a human approves it in four seconds. Daanyaal Bandukwala brings the pharmaceutical commercialization and patient support program lens, and the episode is upfront about the commercial relationship between host and guest, which is why the questioning goes harder than usual. Topics include the Globe and Mail test as a governance heuristic, automation bias in expert decision making, AI disclosure, data residency versus sovereignty, the collapse of the build case in favour of buying, and an honest audit of which governance studies are vendor funded.TakeawaysGovernance is not killing AI projects. Unanswered ownership questions are. Compliance is asking who is accountable, and nobody wrote it down.The Globe and Mail test is the simplest governance heuristic available. If this ended up on the front page tomorrow, how would you look?AI governance lives in onboarding, not in a paragraph buried in a large SOP.Human in the loop is not the same as human accountability. When AI is wrong, expert accuracy collapses, and only personal accountability changes that.A quarterly spot check beats an annual formal audit, and a vendor's reaction to being spot checked tells you more than the audit clause.No independent, non-vendor study isolates governance as the cause of better AI outcomes. It is conviction, not proof, and both host and guest say so.Start with the business problem, not the platform. Evaluate governance before features.Walk out if a vendor claims 100 percent accuracy, says governance can wait, calls integration phase two, or says AI works the same in any industry.Executives are wrong about how their people feel and wrong about what their people are doing, and most are comfortable with unapproved use anyway.Chapters00:00 The Role of Governance in AI03:20 Challenges in AI Adoption19:43 Human-in-the-Loop in AI Practice27:55 Accountability in AI30:48 Responsible AI and Transparency31:58 AI Disclosure and Transparency33:13 AI in Medical Education35:18 Influencer Marketing and AI36:37 AI in Healthcare and Patient Support41:20 Data Resiliency and Sovereignty45:12 The Review Step That Stopped Happening50:07 Regulatory Decisions and Drug Discovery53:11 Build vs. Buy in AI Solutions57:48 Specialized AI Vendors58:30 Why There Is No Good Data on Build vs. Buy1:02:11 How to Choose an AI Vendor1:07:16 Vendor Red Flags and When to Walk Out1:11:52 The Hard Question: Is Governance Actually Proven?1:15:28 Five Beliefs, True or False1:19:18 Shadow AI and the Executive Perception Gap1:23:17 The 95 Percent Pilot Failure Myth1:26:57 Whose Job Gets Smaller1:29:18 Middle Management and Job Security1:32:57 The First 90 Days1:36:01 What Daanyaal Is Least Sure About1:39:57 The Guest Question and Closing