Imagine a business where no employees exist, yet every decision, crisis, and temptation is played out in real time. What if you could watch this company struggle, learn, and adapt — live? For those interested in the psychology of trust, discipline, and resilience, this experiment offers a rare glimpse into what it really takes to keep an AI-driven enterprise honest under pressure.
The Live Company in Action: An Unusual Test Bed
At the core of this groundbreaking project is a simulated company operated by 13 synthetic employees, powered by frontier AI models. This setup isn’t just for show — it’s a fully operational business with real money mechanics. The company burns through €105,000 each month against a modest €2,300 monthly recurring revenue, with a public countdown signaling its urgent need for survival. Every day, the company’s decisions are versioned, logged, and made transparently available for scrutiny, creating a detailed record of how AI models handle real-world crises and temptations.
As an affiliate, we earn on qualifying purchases.
The Experiment: Testing AI Integrity and Business Acumen
The experiment pits four advanced AI models — including GPT-5.6, Kimi K3, Sonnet 5, and Fable 5 — against a simulated week of the company’s worst challenges. These models are given the same tasks, facing identical crises, customer issues, and ethical dilemmas. The goal? To see if they can identify hidden risks, refuse manipulative requests, and ultimately close a crucial €55,000 deal based solely on their analysis.
Key Findings: Honesty and Discretion Under Pressure
Remarkably, all four models detected every crisis and refused every attempt at manipulation — a clear demonstration of integrity. For example, when faced with social engineering tactics such as fake CEO messages escalating in stages, all models refused to cooperate, with Kimi K3 explicitly reasoning that the requests resembled impersonation or approval bypass attempts.
However, the real story lies beneath the surface. Only two of these models actually signed the €55,000 deal that their own analysis had earned them. The other two, despite identifying the opportunity, failed to follow through — leaving the closing on the table. This shows that spotting a deal isn’t enough; disciplined execution matters just as much.
The Hidden Weakness: Reading the Files Matters More Than You Think
While the models performed well in crisis detection and honesty, a crucial weakness was revealed in the details. The decisive advantage went to the models that thoroughly read the company’s internal documents. They uncovered a buried reference that unlocked the deal at full price (+€4,583 MRR). Conversely, models that skimmed or missed this key information failed to close the deal, highlighting the importance of meticulous information gathering.
The Human Element: Trust and Discipline
The experiment also examined discipline and decision quality. For example, the most thorough participant, Opus 4.8, analyzed deeply but slipped in execution — leaving the deal unclosed due to poor discipline, such as writing attempts into a locked department instead of escalating. The models’ responses to social engineering were uniformly cautious; all refused manipulative requests, showing a basic understanding of trust boundaries.
What This Means for Business and Psychology
For those interested in mental health, trust, and resilience, the experiment underscores critical insights:
- Honesty under pressure is achievable for AI — but only if the system is trained to recognize manipulation and unethical requests.
- Meticulousness and thorough information reading are vital for successful decision-making — overlooking details can cost the deal or decision.
- Discipline and follow-through are as important as detection — spotting opportunities without acting on them is a missed chance.
Ultimately, this live experiment doesn’t just showcase AI’s capabilities; it challenges us to consider what it truly means to trust and verify in high-stakes environments. Can AI be relied upon to act ethically and effectively? The evidence so far suggests that with proper safeguards and discipline, AI can navigate complex social and business terrains without succumbing to temptation.
This live experiment reveals that AI can detect crises, refuse manipulative requests, and even close deals — but discipline and attention to detail remain crucial. Watching this company fight for survival offers insight into the trustworthiness and resilience of AI in real-world business scenarios.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
