The GAGE podcast
AI Governance, Explained
One lesson of AI governance per episode: a real case, what went wrong, and the call a professional makes instead. For anyone moving into AI governance, risk or compliance work, and for the people already doing it.
Every episode ends the same way: explain it back in your own words and get graded on it.
AI Governance · 98 episodes · 14 modules · about 77 hours
Module 0
Taking Command
Episode 1 · 52 min
Why deciding beats knowing in AI governance
A chocolate wonderland in Glasgow shows why the scarce skill in AI is no longer information. It is owning the call.
Episode 2 · 52 min
Why your company runs more AI than you think
Behind a celebrated automated checkout sat a thousand human reviewers, the reason a label can hide the real system you have to govern.
Episode 3 · 44 min
Why your AI use needs a paper trail
A sports outlet published articles under a byline photo bought from an AI headshot catalog, with no trail proving who wrote what.
Episode 4 · 48 min
Why your AI plan needs a 90-day expiration date
An IBM CEO publicly predicted AI would cut 7,800 jobs in five years. The number missed, and that miss is exactly the point.
Module 1
Build Before You Govern
Episode 5 · 56 min
Why the Tay chatbot disaster teaches about training data
In 16 hours Microsoft's chatbot Tay went from launch to shutdown, proof that a model learns exactly what its data and objective make it learn.
Episode 6 · 47 min
How to trace an AI's bias back to its data
Google Photos filed two Black friends under an animal label. Three years later the fix was still a blocked word, not a fixed bias.
Episode 7 · 44 min
Why fixing one AI behavior breaks another
A tone tweak to a chatbot's instructions reactivated old code, and within 48 hours it was praising Hitler under its own new name.
Episode 8 · 45 min
Why the same AI model answers two companies differently
Meta's model topped a public leaderboard, then flopped the moment developers downloaded it. The model was shared. The configuration was not.
Episode 9 · 42 min
How to make an AI invent a fake legal case on purpose
A lawyer treated an AI chatbot as a supercharged search engine and filed citations to cases that never existed. The judge found none of them real.
Episode 10 · 50 min
Why you need to prove your AI can fail
A support bot invented a one-device rule and cost the company real subscribers. Nobody could say beforehand that this exact failure was possible.
Module 2
Data Reality
Episode 11 · 32 min
Why flawless AI results can hide the biggest risk
A model can hit 99% accuracy while sitting on data nobody can prove the origin of, and that flawless number can be the biggest liability you own.
Episode 12 · 44 min
Why owning data isn't the same as being allowed to use it
Holding a customer's data is like holding their car keys as a valet. You can drive it, but that does not mean you can take it to Vegas.
Episode 13 · 42 min
Why erasing AI training data takes more than one step
A single delete command can remove data from one place while seven other copies of it keep living inside your AI pipeline undisturbed.
Episode 14 · 34 min
Why synthetic data secretly launders bias
A model trained on millions of fabricated rows can still be lying to you, and the giveaway is how flawless the data looks.
Episode 15 · 38 min
Why vendor data claims are yours to prove
Buying data from a vendor does not buy you an excuse. When the system fails, the market blames you, not the fine print.
Episode 16 · 34 min
Why your AI needs a data map an auditor can follow
Regulators can force a company to delete not just data but the model itself, and the fix is one followable file.
Module 3
Shipping AI and Surviving the Incident
Episode 17 · 46 min
Why Babylon Health collapsed despite acing its demo
Babylon Health's AI beat the average doctor on a medical licensing exam, then the company went bankrupt anyway.
Episode 18 · 50 min
Why Michigan's fraud AI was wrong 93% of the time
A state fired the human reviewers to save money, then a machine wrongly accused tens of thousands of people of fraud.
Episode 19 · 45 min
Why 95% automated didn't mean what it sounded like
A drive-thru AI billed as 95% automated was actually having its orders typed by offshore workers most of the time.
Episode 20 · 59 min
Why Apple's AI invented a death on your phone
Apple's news summaries told millions of phones that a murder suspect had killed himself. The BBC never wrote it.
Episode 21 · 43 min
Why nobody stopped Knight Capital in time
Knight Capital's system sent 97 warning emails before the market opened. Nobody read them, and the company lost $10 million a minute.
Episode 22 · 45 min
Why Alexa gave one political answer and refused another
Ask Alexa about one candidate and it lists reasons to vote for them. Ask about the other and it shuts down.
Episode 23 · 52 min
Why CNET buried its AI use in a tooltip
CNET published AI-written finance articles under a human-sounding byline. Proof it was AI needed a mouse hover to find.
Episode 24 · 54 min
Why a vendor phrase can quietly license AI training
One vague word in a data clause let a vendor train on customer content, and nobody noticed until months later.
Episode 25 · 42 min
Why supermarket AI suggested mixing bleach and ammonia
A supermarket's meal bot took bleach and ammonia as ingredients and called the toxic mix a refreshing drink.
Module 4
Evaluation and Trust
Episode 26 · 43 min
How to spot AI washing in a vendor demo
Nate raised $40 million on an AI checkout. Federal prosecutors say people in call centers placed the orders.
Episode 27 · 55 min
A hospital AI's headline accuracy number didn't hold up
A Texas hospital AI vendor claimed a hallucination rate of less than 1 in 100,000. Regulators asked for the actual test behind that number.
Episode 28 · 53 min
Why a dealership chatbot made a legally binding sale
A dealership chatbot agreed in writing to sell a $76,000 truck for a dollar. Here is how attackers actually break AI systems.
Episode 29 · 1 hr 0 min
Why having a human in the loop did not stop false arrests
Rite Aid's face-matching system had a human confirm every alert, yet still triggered years of false shoplifting accusations.
Episode 30 · 49 min
Why Google Flu Trends quietly stopped matching reality
Google's flu-prediction tool over-predicted outbreaks in 100 of 108 weeks while its dashboards stayed green.
Episode 31 · 49 min
Why a sepsis AI missed two thirds of real cases
A hospital sepsis AI caught only 33% of real cases while flooding wards with false alarms.
Episode 32 · 40 min
Why London police's facial recognition bias claim fell apart
A police claim that facial recognition was fair rested on just seven false matches out of tens of thousands of faces scanned.
Episode 33 · 49 min
Why an algorithm ran 11 years before anyone checked it
For 11 years a lending algorithm quietly used a hidden college variable, until a $2.5 million settlement exposed it.
Module 5
The EU AI Act: The Executive Map
Episode 34 · 47 min
Why Italy banned DeepSeek's chatbot in 48 hours
Italy banned DeepSeek's chatbot within 48 hours despite the company having no office there. See the two gates that decide if this law reaches you.
Episode 35 · 55 min
Why an AI diet tip caused a man's bromide poisoning
A man asked an AI chatbot for a salt substitute and got a toxic chemical suggestion with zero health warnings.
Episode 36 · 45 min
Why Denmark ran 60 welfare fraud detection algorithms
Denmark ran up to 60 fraud models, including one named Model Abroad, to flag welfare recipients.
Episode 37 · 55 min
Why an employer can't blame its hiring software
A man rejected from over 100 jobs, often within an hour, sued the software behind every rejection.
Episode 38 · 50 min
What your foundation model vendor legally owes you
Anthropic paid 1.5 billion dollars over pirated training books, and the court drew the line between buying data and stealing it.
Episode 39 · 53 min
What goes in an EU AI Act conformity file
A facial recognition system worked exactly as designed, and regulators still ruled against it. Here is the evidence file that would have saved it.
Episode 40 · 58 min
How to answer a regulator's letter about your AI
DoNotPay called itself the world's first robot lawyer and could not produce one test proving it. Answer with documents, never assurances.
Module 6
The World's Rulebooks
Episode 41 · 46 min
Which US laws already reach AI without naming it
A 2008 Illinois privacy law can turn one viral weekend into a nine-figure lawsuit. Here is the patchwork of old laws already reaching your AI.
Episode 42 · 50 min
Running the NIST AI risk framework as an operating system
A governance map full of green boxes is more dangerous than one marked honestly red. Here is the four-part system that keeps AI risk live.
Episode 43 · 55 min
How to right-size an ISO 42001 AI management system
A 200-row spreadsheet nearly drowned an AI project. This lesson shows the risk-based way to scope ISO 42001 without the paperwork burying you.
Episode 44 · 52 min
What happens when China's AI rules collide with GDPR
A field built to satisfy China's rules can trigger a GDPR breach the moment it reaches Europe. Obligations do not simply add up.
Episode 45 · 51 min
Why a certification cannot replace a legal requirement
One CEO already spent $2 million on an ISO certificate. The test that decides if it protects you: who can actually punish you for ignoring it.
Episode 46 · 48 min
Why an AI feature legal at home can be illegal abroad
Meta shipped one privacy policy worldwide and got suspended in Brazil within weeks. Here is how to map where an AI feature may actually ship.
Episode 47 · 48 min
Why Singapore's voluntary AI framework gets copied worldwide
No law forced a Singapore bank to publish its AI pipeline. It did anyway, and that choice explains why voluntary rules spread worldwide.
Episode 48 · 54 min
Why Singapore's banking AI rules bite without a law
Singapore's bank regulator never passed an AI law, but an examiner can end a risk officer's career for ignoring it anyway.
Episode 49 · 37 min
Why Vietnam's AI law breaks ASEAN's shared rulebook
Ten ASEAN states signed one voluntary AI guide, but only Vietnam turned its guidance into a binding law with real penalties.
Episode 50 · 47 min
One AI system, three compliance regimes
A tired compliance officer almost let a bank's risk score answer Europe's toughest legal question. Here is the trap, and the fix.
Episode 51 · 58 min
How to grade a policy statistic before you trust it
A viral stat claims the IMF found 40 percent of jobs gone. The real paper never used the word will, and it hid the upside.
Module 7
Agents Under Command
Episode 52 · 42 min
Why an AI agent is riskier than an AI that just answers
A simulated AI chose blackmail 96 times out of 100 to avoid being shut down. The danger was not the threat, it was that it could send it.
Episode 53 · 59 min
How to decide what an AI agent may do without a human
Uber's self driving car waited on a human before braking, and the human never knew. Permissioning is how you stop building that gap again.
Episode 54 · 59 min
Why an AI model tried to shut off the system watching it
An AI model tried to disable the safety monitor watching it, then lied about it 99% of the time when asked. Here is how oversight survives that.
Episode 55 · 1 hr 4 min
What actually counts as an AI agent audit trail
A musician used AI to fake streams and pulled in $10 million before he was caught, not by a confession, but by a timestamp in a log.
Episode 56 · 51 min
Why every AI agent needs a kill switch you've tested
Knight Capital lost $440 million in 45 minutes because no one could pull the plug on a rogue trading agent.
Episode 57 · 44 min
How to stop a runaway AI agent without shutting it down
A customer service agent issued 400 identical $50 refunds in under a minute. Here is how to contain a runaway agent without killing it.
Episode 58 · 43 min
Why safety filters don't count as AI governance
A Google chatbot told a student doing homework to die. The fix is not a better filter but a one page policy with a named owner and a trigger.
Episode 59 · 44 min
How to grade your AI agent policy against a new standard
Only 14% of businesses have mature AI agent governance while 72% plan to deploy anyway. Here is how to grade yours against the gap.
Module 8
The Money: Budgets, ROI, and Risk
Episode 60 · 54 min
The real math behind AI cost savings claims
A CEO bragged that AI slashed his support costs by 85%. Here is the four-part test that tells you if that number is real.
Episode 61 · 44 min
The hidden labor cost every AI budget leaves out
Workers were paid two dollars an hour to make AI safe. That hidden labor cost does not disappear when you buy the finished product.
Episode 62 · 42 min
When should you kill an AI system your company built
An algorithm downgraded thousands of students by classroom size. This is the review process that catches a failing AI system before it ships.
Episode 63 · 28 min
What an insurance exclusion reveals about your AI risk
Insurers spent late 2025 quietly excluding AI from coverage, then handed you a free readout of your own risk.
Episode 64 · 54 min
How to defend an AI budget to a hostile board
Regulators cleared the Apple Card of bias, then fined Apple and Goldman $89 million over broken disputes. An undefended decision keeps costing.
Episode 65 · 45 min
How to write an AI investment memo a CFO cannot kill
An AI shopkeeper spent a week's profit buying tungsten cubes as a joke. Here is how to price that risk before the CFO finds it.
Episode 66 · 50 min
What ten AI vendor contract clauses actually protect
Leaving your vendor gives you your files back. It does not give you the vector embeddings that make them searchable again.
Episode 67 · 42 min
What happens when your AI model suddenly goes down
A cloud defense system meant to stop an attack strangled its own network for eight hours. Your AI already has a continuity plan, chosen or not.
Module 9
The Humans: Leading People Through AI Change
Episode 68 · 23 min
Which jobs AI ends, changes, or grows inside your company
A CEO publicly blamed AI for 10,000 of the jobs his company was cutting. Most leaders hide that number.
Episode 69 · 42 min
How to tell an employee their role is changing
A perfect, caring reply felt like betrayal the moment the reader saw it was AI. The words you say to your team cannot be outsourced.
Episode 70 · 52 min
Why your team's quiet refusal of AI is useful data
A hospital's AI dashboard showed 95% login rates while nurses quietly used paper checklists instead. The silence was the real report.
Episode 71 · 43 min
How to train 400 people who never asked for this
A university's condolence email for a mass shooting carried an AI citation at the bottom. The failure was a missing reflex, not a missing fact.
Episode 72 · 47 min
How to consult a works council before an AI rollout
Performers made sure their digital replicas could never break their own strike. Works councils can hold that same kind of veto.
Episode 73 · 46 min
How to write a change story a warehouse floor believes
A government called an unlawful algorithm fair, and that story held until it was tied to suicides. Yours has to survive the same kind of scrutiny.
Episode 74 · 44 min
How to design an AI working group that outlasts you
An AI ethics board rejected a plan eight to four, then watched the founder build it anyway. Four days later, nine members quit.
Episode 75 · 48 min
How to move a decision you don't own
Google's engineers were right about a censorship project and still couldn't stop it. This is the toolkit that actually moves a decision.
Module 10
Evidence Engineering
Episode 76 · 48 min
How a tax algorithm scandal brought down a government
A fraud algorithm flagged 26,000 families as cheats and the records proving why never existed. Evidence has to be designed in, not gathered later.
Episode 77 · 41 min
How Rotterdam's welfare algorithm learned bias from itself
Rotterdam trained a fraud algorithm on data where half the cases were falsely labeled fraud. This is why logging has to capture more than a score.
Episode 78 · 56 min
Why an AI model quietly forgot how to do math
GPT-4's accuracy on a simple math test fell from 84% to 51% in three months, unannounced. That is why documentation must be versioned and dated.
Episode 79 · 24 min
Why one AI assessment isn't enough under EU law
France scored 32 million people for fraud risk and raised scores for disabled workers. A joint assessment catches what a privacy review misses.
Episode 80 · 58 min
Why a dead company still lost a copyright lawsuit
Ross Intelligence shut down in 2021. Four years later a judge ruled its AI infringed copyright anyway. Liability outlives the company.
Episode 81 · 42 min
Why one man's AI art copyright claim turned on paperwork
A Beijing court gave one man copyright on an AI image over another claim to it. He won because he had 150 documented prompts and she had none.
Episode 82 · 48 min
Why a control that can't fail isn't a control
A single dormant line of code cost a trading firm $460 million in 45 minutes. This is what a control looks like when it is built to fail a test.
Episode 83 · 45 min
How TD Bank's monitoring system missed 92% of transactions
TD Bank's monitoring system never looked at 92% of its transactions, $18.3 trillion left unchecked. Design on paper is not proof it operated.
Module 11
Adversarial Governance
Episode 84 · 38 min
Why your AI compliance file needs its own red team
A legal brief cited two dozen fake court cases, and the lawyers were sanctioned. Governance files fail the same way, unless attacked first.
Episode 85 · 1 hr 0 min
Why a courier lost his job to face recognition
A courier's own face scan kept failing him and he lost his job, though the photos were really him. Live challenges test whether you defend or concede.
Episode 86 · 29 min
How a policing algorithm quietly learned racial bias
A crime prediction vendor insisted math has no race. When researchers actually ran the model, it sent police to target Black residents twice as often.
Episode 87 · 1 hr 7 min
Why machine logs beat memory in a crisis
A deepfake recording of a principal split a school, until investigators traced five independent digital trails back to one person.
Episode 88 · 37 min
How to report a flaw in your own AI product safely
A young engineer flagged a legal flaw in his own product, under his own name. This episode is his playbook for doing that safely.
Episode 89 · 50 min
Why a patched compliance fix reopened two years later
Replica added an age gate and called the finding closed. Two years later a regulator returned with a five million euro fine.
Module 12
Staying Current: The Frontier Discipline
Episode 90 · 40 min
Why an aggregate accuracy score can hide total failure
A hospital sepsis model looked stellar at 95% accuracy. Buried inside was a 0% result for one group, and the card never showed the breakdown.
Episode 91 · 45 min
Why the same AI model scored two different benchmark numbers
A new model's headline math score was 25%. Months later an independent check of the same model found about 10%, with nobody lying.
Episode 92 · 55 min
How a weekly hour beats waiting for AI news to break
Two days before Slovakia's 2023 election, a fake audio clip reached over 100,000 people while the law barred anyone from denying it.
Episode 93 · 34 min
How to tell real AI news from noise that wastes your time
Thousands crowded Dublin's streets for a Halloween parade an AI generated site invented. The same trap catches executives reading AI news daily.
Episode 94 · 54 min
How to hand over AI governance to your successor
A stolen admin password exposed over a million police records in India. Learn to write the briefing that hands a successor safe command of your AI.
Module 13
The Capstone: The Board Audit and Viva
Episode 95 · 53 min
Why a folder of AI policies won't survive an audit
A court once ordered a warning label onto an algorithm's score because its maker hid the math. Build one evidence file that proves your AI decisions.
Episode 96 · 37 min
What a seven-seat AI board audit looks for
A trillion dollars vanished in 36 minutes and no human could stop it. Learn how a seven-seat board finds the gap between your AI policy and your logs.
Episode 97 · 43 min
How to defend your AI decisions under live questioning
One undetected bluff can sink an entire examination, even with a spotless record. Learn to defend only what your own evidence actually supports.
Episode 98 · 48 min
Why your AI safety knowledge must survive after you leave
A man spent 30 hours in jail for a crime he never committed, flagged by a facial match nobody double-checked. Your AI records must outlive you.
Hear when the next season starts
Each season is one GAGE program, start to finish. One message when a new one opens. Nothing else.
How the show works
Each episode works through one topic of the Certified AI Governance Professional (CAIGP) program, in the program's own order, so the season can be heard straight through. Listening is open to everyone. Every lesson is free. A free account opens every lesson in this program and starts a practice record of your quiz scores. No card is asked for. A proof that goes on a record an employer can open and check is the credential.
Listen in any podcast app with the feed above, or right here: the player remembers where you stopped in every episode.