01
Accountability erodes
Decisions attributed to 'the model' are decisions no human owns. Board minutes fill with passive voice.
AMPLIFI AI LEADER COMPASS
To believe
or not
to believe...
Automation bias is the quiet tendency of capable people to over-trust the output of a system — a dashboard, an AI model, a vibe-coded prototype — and to under-weight the evidence in front of them. It is now one of the most under-managed risk in the modern workplace.
§ 01
The definition
Automation bias is the tendency to favour suggestions from an automated system — and to disregard contradictory information from a human source — even when the automated system is demonstrably wrong.
Errors of commission
You do what the system tells you to do, even when your own information should have stopped you.
Errors of omission
You fail to notice a problem because the system didn't flag it. The absence of an alert is treated as an all-clear.
"The more reliable the automation appears, the less vigilant its user becomes. Trust and attention move in opposite directions."
§ 02 · Governance
What it does to your governance stack.
02
Audit trails collapse
An LLM's reasoning is not a record. Without provenance, regulators, insurers and juries cannot reconstruct why a call was made.
03
Risk concentration
When every team uses the same three models, correlated errors become systemic. Diversity of thought is quietly replaced by diversity of prompt.
04
Prototype gravity
A vibe-coded prototype anchors the roadmap. Discussion narrows to what the prototype already does, not what the problem actually needs.
05
Deskilling
The junior analyst who never had to work the model by hand cannot spot when the model is off. Expertise atrophies within one cohort.
06
Regulatory exposure
EU AI Act (Article 4), ICO, FCA, DIST — all expect documented human oversight. 'The tool said so' is not a defence.
§ 03 · Cognition
The invisible tax on your people.
A workforce that stops thinking is a workforce that stops learning. Automation bias taxes creativity, curiosity and craft — and the invoice arrives late.
- Divergent thinking
- When the first draft comes from a model, the second, third and fourth ideas rarely happen. The team converges on the machine's midpoint.
- Metacognition
- Employees stop asking 'how do I know this?' The muscle of reflective judgement weakens through disuse.
- Domain fluency
- The tacit knowledge that used to come from wrestling with a problem is outsourced to a model that has none.
- Psychological ownership
- Work you didn't author is work you will find challenging to defend. Engagement follows authorship.
§ 04 · Diagnostic
Did you unwittingly
hand the keys over to themachine?
Ten questions. Six minutes. A ruthless read on how much of your judgement you have already handed over — and where to install the guardrails first.
§ 05 · Next steps for leaders
Turn awareness into action.
Automation bias is a leadership risk, not an IT risk. Choose one action from below and make it part of your next team meeting.
01→
Diagnose your bias
Take the 10-question self-assessment to see how automation bias shows up in your own decisions and where your blind spots are.
Start the quiz
02→
Equip your teams
Share the employee field guide with your team so they can spot LLM-generated reports and vibe-coded prototypes that narrow the conversation.
Read the field guide
03→
Install guardrails
Use the printable checklist to put human oversight, provenance checks and escalation rules in place before the next critical decision.
Get the checklist
Want to go deeper? Run the diagnostic with your team and compare scores? Or bring the team together to build your team AI strategy, use policy and governance. The conversations we start are often more valuable than the scores themselves.
BOOK A FREE 30 MINUTE CONSULTATION CALL→Free · Leader toolkit
Get the guardrails checklist and participant workbook.
One email. Two field-tested resources for you and your team. No spam — unsubscribe any time.
Prefer the full form? Open the toolkit page →