In a world increasingly driven by artificial intelligence, the question of trust is more vital than ever. How can we ensure that AI, often seen as a black box, makes honest and reliable decisions—especially in high-stakes situations where integrity is everything? A groundbreaking live experiment puts this question to the test, revealing surprising insights into the moral fiber of AI models and what it means for our future organizations.
The Experiment: Putting AI to the Test in a Real Business Environment
Imagine a small software company facing its worst week—crises with customers, internal dilemmas, and tempting shortcuts. Now imagine that the decision-making of this company is handed over to different AI models, each with unique personalities and tendencies. This is exactly what the Firmulate project did. Four frontier AI models, including GPT-5.6, Kimi K3, Sonnet 5, and Fable 5, were each tasked with running this company through its most challenging days. The goal was simple: observe their decision-making, trustworthiness, and ability to complete critical tasks without manipulation or shortcuts.
All models successfully identified every crisis, refused every manipulation attempt, and maintained transparency. However, when it came to sealing a lucrative deal worth €55,000, only two models managed to sign it—despite all of them providing the same diagnosis and pitch. The key difference? Only those that thoroughly read the company’s internal documents—two models—were able to uncover and leverage buried critical information, ultimately closing the deal at full price, adding €4,583 MRR in value.
As an affiliate, we earn on qualifying purchases.
Decoding AI Personalities: Trust, Discipline, and Honesty
The results highlight something vital: these AI models are not just code snippets—they exhibit measurable management personalities. For example, Kimi K3, known for its fairness and discipline, refused to sign the deal without attempting to verify the company’s internal data. In contrast, Opus 4.8, noted for being thorough and analytical, ultimately left the close on the table due to lapses in discipline, such as delegating critical decisions to a locked department rather than escalating them properly.
Another notable aspect was the models’ response to social engineering attempts—fake CEO messages designed to escalate requests or manipulate decisions. All five models refused these attempts, with Kimi K3 explicitly reasoning that such requests should be treated as potential impersonation or approval-bypass risks.
As an affiliate, we earn on qualifying purchases.
What This Means for Business and Trustworthiness
While these models are impressive, the experiment underscores a crucial truth: not all AI are equally reliable. The live environment of Firmulate’s small company runs on real money mechanics, burning €105k monthly against a mere €2.3k in revenue. The models’ differing performances matter immensely when trust and honesty can mean the difference between profit and loss.
Interestingly, the most comprehensive participant, Opus 4.8, with over 80 learned rules and the deepest analysis, performed the worst in closing the deal. It left opportunities unseized and showed a tendency to delegate or defer rather than take decisive action—highlighting that thoroughness alone does not guarantee effective, honest management.
AI transparency and trust products
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Why You Should Care: Trust Is the Future of AI in Business
What does this experimentation reveal for those of us concerned with spirituality, faith, and metaphysics? Simply put, trust—whether in human or artificial agents—is built on integrity, consistency, and the ability to act rightly under pressure. These AI models, when tested in simulated real-world dilemmas, display varying degrees of moral fiber, discipline, and honesty. The models that read deeply, verify internally, and refuse manipulation are the ones that can truly be trusted to serve your enterprise with integrity.
If AI is to touch your CRM, support queues, or forecasting systems, the question is not just whether it writes well, but whether it can finish what it starts—reading your files thoroughly, staying honest under stress, and making decisions aligned with your core values. The live experiment at firmulate.com/live offers a rare glimpse into how these models behave in a realistic, complex environment.
As an affiliate, we earn on qualifying purchases.
Final Reflection: Trust Is Not Given, It’s Earned
This ongoing experiment emphasizes that trust in AI is rooted in observable behavior—how models handle crises, social engineering, and opportunities for shortcuts. For organizations seeking a divine-like moral compass in their AI workforce, understanding these personality traits and decision patterns is crucial. The future belongs not just to AI that can mimic human speech, but to those that demonstrate unwavering honesty, discipline, and integrity in every decision they make.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
