Pro AI · v3.0 · New
A careful read on Machiavellianism.
With the safeguards on the surface, not buried.
Every Speaker Insights profile now includes a Mach-IV inline card — a calibrated read on whether someone treats relationships as strategic resources or intrinsically valuable. Three factors. Conservative bands. The limitations stated where you can see them, not in a footnote.
Why include a charged framework at all.
Christie & Geis published the Mach-IV in 1970 to measure a personality construct they called Machiavellianism — a strategic, instrumental, cynical interpersonal style. Five decades and thousands of peer-reviewed citations later, it's one of the most consistently-validated scales in organizational and social psychology.
The framework matters because some of the most useful coaching advice depends on knowing whether the person you're about to talk with treats relationships as a means to an end. A Bonfiyah profile that includes a calibrated Mach read shapes the "how to work with this person" synthesis in concrete ways: don't expect them to disclose unilaterally; treat their handshake as the contract, not the relationship; protect yourself in writing.
It's also, more sharply, a self-protective tool. Combined with low Big Five Agreeableness and a Belbin Shaper profile, high Mach is the third leg of what organizational psychology calls the Dark Triad — narcissism, Machiavellianism, psychopathy. Most people will never have a Dark Triad colleague. Some people work with one for years before they notice. The framework is in Bonfiyah for the same reason it's in the literature: noticing earlier matters.
The three factors
Tactics. Views. Morality.
The published Mach-IV produces a single composite. Hunter, Gerbing & Boster (1982) and Williams, Paulhus & Hare (2009) cleanly separated three underlying factors that we score independently and average to a composite — so you can see which dimension is driving the read.
Factor 1
Tactics
Willingness to manipulate. Instrumental framing of other people.
Lexical signals: instrumental verbs ("leverage," "play," "get them to"), persuasion-as-manipulation framing, the language of moves rather than conversations.
Factor 2
Views
A cynical worldview. Distrust of others' motives.
Lexical signals: zero-sum framing ("everyone's looking out for themselves"), distrust phrases ("you can't trust anyone in this"), expectation that interlocutors are pursuing hidden agendas by default.
Factor 3
Morality
Moral relativism. Ends-justify-means thinking.
Lexical signals: ends-justify-means framing ("whatever it takes," "results matter, not how"), rules-don't-apply phrasing, language that treats norms as obstacles rather than commitments.
Composite
The composite Mach score is the equal-weight mean of the three factors, mapped to a 0–1 scale with the population mean anchored at 0.50. Above 0.65 reads as "high Mach." Below 0.35 reads as "low Mach." Anything in the middle reads as "Population-typical" — see the next section on why we render moderate scores as words rather than numbers.
How the card reads in the app
Below 0.35
Low Mach
"Reads below population-typical on the Mach-IV composite."
Treats relationships as intrinsically valuable. Tends to over-disclose, expects others to do the same, can be slow to recognize when someone is operating in a different mode.
0.35 to 0.65
Population-typical
"Mach-IV composite reads near population centre."
We deliberately do not render a numeric score in this band. The literature does not support a confident high-vs-low call from a small text sample inside this range, and "borderline Machiavellian" is not a label any honest framework should hand you.
Above 0.65
High Mach
"Reads above population-typical on the Mach-IV composite."
Treats relationships as strategic resources. Coaching advice shifts: don't expect unilateral disclosure; protect yourself in writing; treat their handshake as the contract.
Safeguards on the card itself
Five guardrails, surfaced where you'll see them.
1. Confidence is capped at medium. Even with abundant quotes, Bonfiyah will not report "high confidence" on a Mach-IV read derived from speech. Text patterns are not the ground truth — behavior is — and we don't pretend otherwise.
2. Moderate scores render as words, not numbers. A composite in the 0.35–0.65 band reads as "Population-typical," not as "57th percentile." The line in the literature where a confident call can be made is not in the middle of the distribution, so we don't pretend the middle of the distribution is informative.
3. Occupational vocabulary is flagged inline. Sales, negotiation, executive-coaching, and litigation contexts produce instrumental vocabulary that elevates the Tactics factor independently of trait Machiavellianism. The card calls this out so the reader doesn't conflate professional register with personality.
4. Minimum sample of twelve quotes. Below twelve, the card shows a placeholder asking for more recordings before issuing any read at all.
5. Methodology and limitations always one tap away. The card always surfaces a link to the methodology + limitations text. The user is never expected to trust the bottom-line label without seeing the work.
What Bonfiyah's Mach-IV is — and isn't — for.
It is for: helping you understand the people you personally talk with, in your own conversations. A board director making sense of a deal counterparty's posture. A coach orienting around a client's worldview. A negotiator preparing for a difficult counterparty. A spouse processing a difficult family member. A founder reading the room before a hard board meeting.
It is not for: hiring decisions, promotion decisions, performance ratings, custody evaluations, security clearances, or any other consequential decision about another person. The Colorado AI Act's high-risk classification for AI systems making consequential decisions in employment, and the EU AI Act's parallel framework, both apply to anyone using personality-inference AI in those contexts. Don't use Bonfiyah's personality reads as inputs to consequential decisions about other people without your own legal review and your own ground-truth instruments.
Mach-IV in its native form is a forced-choice self-report instrument completed by the person being measured. Bonfiyah's read is a conversation-pattern proxy. The two are not interchangeable, and we say so on the card, in the methodology, and here.
Privacy & deletion.
A Mach-IV read is part of the Speaker Insights profile for a given person. It lives in your account on our backend so it can synthesize across recordings; it is isolated from every other Bonfiyah user's library. The person being read does not have access to it unless you choose to share it with them — and Bonfiyah supports a one-page Speaker Insights export specifically so you can hand the read to the person it describes if asked.
Delete the speaker, and the framework analyses go with them. Delete your account, and the whole library cascades. The source audio that fed the inference is on the standard 7-day auto-purge regardless.
FAQ
What is Mach-IV and where does it come from?
The Mach-IV is a 20-item Likert scale published by Christie & Geis in 1970 (Studies in Machiavellianism, Academic Press). It measures a personality construct called Machiavellianism — a strategic, instrumental, cynical interpersonal style. One of the most widely used personality instruments in organizational and social psychology research, with thousands of citations.
What are the three factors?
Hunter, Gerbing & Boster (1982) and Williams, Paulhus & Hare (2009) identified three underlying factors that Bonfiyah scores independently. Tactics — willingness to manipulate. Views — cynical worldview. Morality — ends-justify-means thinking. The composite is the equal-weight mean.
How does Bonfiyah avoid mislabeling someone as Machiavellian?
Five guardrails: confidence capped at medium; moderate scores render as "Population-typical" not as a number; occupational vocabulary flagged inline; minimum twelve quotes before any read; methodology and limitations always one tap away.
Is this a substitute for the actual Mach-IV questionnaire?
No. Mach-IV in its native form is a forced-choice self-report. A speech-derived inference is a conversation-pattern proxy — useful as a hypothesis, not as a verdict. We say so on the card, in the methodology, and on this page.
Could this be used in hiring or performance reviews?
We strongly recommend against it. The Colorado AI Act (effective Feb 2026) and the EU AI Act both treat AI systems making consequential decisions in employment as high-risk and impose specific obligations on developers and deployers. Bonfiyah's personality reads are personal-context tools — to help you understand the people you talk with — not inputs to consequential decisions about other people. If you intend to use them that way, do your own legal review first.
Tell me when Mach-IV ships its next refinement
We're iterating the lexicons, the band thresholds, and the cross-framework triangulation with Big Five Agreeableness. Subscribe if you want the changelog when those land.
No spam. We use ConvertKit. See our privacy policy.