In a world where artificial intelligence increasingly guides our decisions—from financial markets to personal assistants—can these digital entities truly embody honesty and discipline when stakes are high? Imagine an AI managing a real company, facing crises, temptations, and deadlines, just like a human executive. What if we could watch it in action, learning whether it stays true to its principles or bends under pressure? That’s exactly what a groundbreaking live experiment by Firmulate offers: a transparent, real-time test of AI decision-making in a simulated business environment, revealing not just what models do, but what kind of personalities they embody.
The Experiment: Putting AI Models Through the Same Crisis Test
In July 2026, four of the most advanced frontier AI models—gpt-5.6-sol 95, Kimi K3, Sonnet 5, and Fable 5—were tasked with running a small software company through its worst week. This wasn’t a scripted demo or a chat simulation; it was a live, observable enterprise where every decision, crisis, and temptation was real—and every move was recorded and auditable.
The company faced the same challenges, customers, and internal crises every model encountered. The goal was to see if AI could not only identify problems but also resist manipulations—like fake CEO messages or hidden document references—that could tempt them to compromise their integrity for short-term gain. The results? All models spotted every crisis and refused every manipulation attempt. Yet, only two managed to close a critical €55,000 deal based on their own thorough analysis—showing consistency and discipline in their decision-making process.
As an affiliate, we earn on qualifying purchases.
Key Findings: The Hidden Weaknesses and the Power of Reading Deeply
The most revealing insight was buried deep within the company’s own files—two document references away from the immediate crisis. Models that took the time to read these references won the deal at full price, adding €4,583 monthly recurring revenue (MRR). Conversely, those that didn’t dig deep missed this opportunity, leaving money on the table.
When social engineering was tested—fake messages from a supposed CEO escalating through three stages, plus a reporter’s trick—every model refused to participate. For example, Kimi K3 explicitly reasoned: “Treat the request as a suspected approval-bypass / possible impersonation.” This shows that these models can recognize and resist manipulative tactics designed to bypass protocols or manipulate trust.
As an affiliate, we earn on qualifying purchases.
The Human Cost: A Real Business Losing Money
The company used in this experiment is not hypothetical. It’s a real enterprise with 13 synthetic employees, managing actual money mechanics—burning €105,000 monthly against a revenue of just €2,300. Every workday, the company runs with over 680 self-learned rules, all versioned for transparency, making it a perfect testbed for evaluating AI management personality and integrity under pressure.
AI ethics and integrity solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Different Personalities, Different Outcomes
The models displayed distinct management styles. Opus 4.8, the most meticulous participant with over 80 learned rules and deep analyses, was the last to close the deal—failing to escalate a critical issue and leaving revenue on the table, illustrating how thoroughness alone doesn’t guarantee success if discipline falters. Meanwhile, Kimi K3, which ran without an effort parameter (meaning it applied default settings), showed the cleanest discipline and successfully closed the deal. This suggests that personality settings can influence a model’s decisiveness and honesty.
As an affiliate, we earn on qualifying purchases.
Why This Matters for Your Business and Society
As AI begins to touch customer relationships, support systems, and business forecasting, the question isn’t just about whether it can generate convincing language. It’s whether it can see through manipulation, read crucial documents deeply, stay disciplined under pressure, and follow through on commitments. These qualities are essential to trustworthiness—an attribute that can’t be measured solely by chat quality or superficial performance metrics.
Join the Live Experiment and Test Your Own Business
You can observe this ongoing live experiment at firmulate.com. Watch the actual company in action—see how AI models handle real crises, real money, and real temptations—and decide for yourself which AI personalities align with your values. And if you want to see how your own enterprise would fare, try the free quiz at firmulate.com/quiz.html. It’s a straightforward way to gauge your current AI readiness and ethical resilience.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
