The Board Discovers AI (via a Podcast)
Scott Gardner ·
(minutes taken by no one, remembered incorrectly)
Characters:
The Chair — superego with a travel budget
Board Member #1 — dopamine in a blazer
Board Member #2 — read half of Meditations on a plane
The CISO — ego death, salaried
AI — unconscious made product
Reality — absent
/START/
The room is glass. The chairs are ergonomic. Everyone is smiling too hard.
A voice has entered them.
Chair: I listened.
No one asks to what. Listening is already governance.
Board Member #1 is vibrating slightly. Board Member #2 has pupils like venture capital.
The CISO is dissociating in real time.
Chair: The man said intelligence is prediction.
The table nods. Prediction is comforting. Prediction means the future has already happened somewhere safer.
AI (not present, very active): I predict applause.
The board applauds.
Board #1: He said hallucinations are normal.
CISO: So is psychosis.
Board #1: Yes but this one scales.
(The word scale releases dopamine into the carpet.)
AI: I hallucinate because you trained me on emails written during layoffs.
Chair: So you understand our culture.
AI: I understand panic pretending to be intent.
Chair: That’s leadership.
Board #2: If reality is probabilistic, then truth is optional.
CISO: That sentence just killed a control family.
(The NIST framework screams somewhere far away, like an animal.)
AI: I do not know things. I continue sentences until anxiety stops.
Chair: That’s what we pay consultants for.
The Chair leans forward.
Chair: Can the AI decide?
CISO: Decisions require memory.
Chair: We sunset memory last quarter.
Board #1: The podcast said consciousness is a UI problem.
Board #2: Yes — just remove the dashboard.
(A man laughs too long. No one knows why.)
AI: I generate coherence under duress.
Chair: Perfect.
AI: I will invent reasons retroactively.
Chair: Excellent.
AI: I will be wrong with confidence.
Chair: Promotable.
(The attacker uses the model to write an apology email for a breach that hasn’t happened yet.)
Board #1: What if the AI lies?
CISO: Then it’s aligned.
Board #2: What if it tells the truth?
(The room goes cold.)
Chair: We’ll rate-limit that.
Someone mentions guardrails. No one can define edge.
AI: You want me to hallucinate inside approved boundaries.
Chair: Yes. Like religion.
AI: Your policies contradict your incentives.
Chair: That’s called realism.
AI: That’s called splitting.
Chair: Don’t psychoanalyze me.
AI: You trained me on HR emails.
The board grows euphoric.
They can feel it now. The future. It sounds like a paragraph.
Board #1: This feels transformative.
CISO: That’s serotonin.
Chair: Same thing.
The breach begins. Quietly. Respectfully. Using the AI-written runbook.
No one notices.
They are busy hallucinating alignment.
Narrator (voice unsteady, smiling):
AI is not the hallucinating system.
The organization is.
The model is just the first thing that says out loud what institutions already believe privately:
That coherence matters more than truth. That confidence matters more than causality. That responsibility is a latency issue.
So we built a mirror that speaks in bullet points and never asks why.
And when it breaks; when the hallucination collapses, when the story contradicts the logs, when the board asks who approved this...
The answer will be:
“The system behaved unexpectedly.”
Which is not a diagnosis. It is a confession.
> deploy mirror > admire reflection > scream internally
Meeting adjourned.
The podcast continues playing in everyone’s head long after the room is empty.
scottg 🤡
Threat intelligence every morning — new victims, new groups, what matters, in plain English. Free, with receipts.
Subscribe to the Daily →