Lack Of Accountability
The system cannot be held responsible. Someone will be. The only open question is who.
This page covers one specific way AI gets things wrong at work, and what to do about it.
It runs in order. What goes wrong, why it happens, where you'd notice it on an ordinary day, who takes the blame, roughly what it costs, and the check that catches it. Then one thing to try this week.
The dollar figures are estimates, not measurements. The assumptions behind each one are printed right there, so you can swap in numbers that fit your job. Anything actually measured carries an OBSERVED tag.
What is lack of accountability?
Not technical. Structural — and it is the reason this book exists.
An AI deployment moves work from a person to a system. It does not move accountability, because accountability is assigned by contract, regulation, and organizational hierarchy, none of which recognize a model as a party. So a gap opens: the work moved, the responsibility didn't.
The gap does not stay open. It collapses onto the nearest human — and "nearest" means last to touch the file, not most senior. Risk transfers downward through the chain: vendor disclaims to buyer, executive delegates to manager, manager assigns to individual contributor.
At every step, authority decreases and exposure increases. The person with the least ability to change the system carries the most consequence for its errors. That is not a bug in how organizations adopt AI. It is the default outcome, and it happens absent deliberate intervention.
What do people assume?
That when an AI-assisted process produces harm, responsibility attaches somewhere proportionate — that vendor, tool, or "the system" absorbs a meaningful share.
Liability follows the human decision-maker. Vendor terms disclaim consequential damages. The organization owns the output, and inside the organization, ownership descends to whoever last touched it.
This is the mode that makes the other twenty-three matter.
Where does it show up at work?
A regulated communication is drafted with AI assistance, reviewed quickly, and sent. It contains a statement that violates a disclosure requirement.
The vendor's terms disclaim liability. The executive who mandated AI adoption for efficiency is not named. The manager who approved it says it was reviewed. The specialist who sent it is the one in the meeting with compliance.
Who carries the downside?
That is the entire mode. Trace it every time:
Vendor → contractually insulated Executive → set the mandate, absorbed none of the risk Manager → delegated review, retains plausible distance You → name on the record, in the room
What does it cost?
[MODELED — not reported]
ASSUMPTIONS
Career cost of being the named party
in a material AI-attributed failure:
- Lost promotion cycle: $15,000 – $40,000
- Bonus / rating impact: $5,000 – $25,000
- Involuntary transition risk: variable, severe
Annualized exposure: personal, not organizational — and uninsurable
This is the only mode in Part II where the exposure is yours rather than the company's. That is precisely why it's worth building a control for on your own initiative, without waiting for a budget.
How do you control for it?
Documented decision ownership. Every consequential AI-assisted output carries a record: what was verified, by whom, against what source, on what date, and what was explicitly not verified.
The record does not eliminate responsibility. It converts an undefined, unlimited exposure into a defined and defensible one. "I did not check that" with a documented scope is a survivable position. "I don't know what was checked" is not.
CONTROL COST Consequential outputs: 250 / year Record time: 3 minutes each Annual: 12.5 hours Fully loaded rate: $70 / hour
Annualized control cost: $875
What should you do this week?
RECOMMENDATION
Start the AI Error Log (Part V) this week. Not because errors are coming — because the log is the evidence that you were the person exercising judgment.
It serves three functions simultaneously, which is why it's the single artifact this entire book is organized around:
1. Protection. A documented scope of what you verified. 2. Evidence. Material for performance reviews, promotion cases, and interviews. 3. Leverage. Data nobody else in your organization has, about a risk everyone acknowledges and nobody has measured.
Part IV turns this log into a ninety-day record. Part V gives you the template. Start it before you finish the book.
Evidence
REPORTED Organizations held responsible for the output of their AI systems in customer-facing contexts.
ANALYSIS The downward risk-transfer chain — authority decreasing while exposure increases — is the author's structural argument and the central thesis of the book. Label it as analysis. It is more persuasive labeled honestly than dressed as a finding, and mislabeling the book's own thesis would be the most damaging possible failure of its own standard.
This is one of 24 failure modes. The book gives you all of them — plus the controls that catch each one and a 90-day plan to prove you ran them.
Preorder the Book