Imagine having an AI leader who handles your company’s toughest week—making decisions under pressure, reading critical files, and refusing to be manipulated. Now, what if you could see which AI models are truly trustworthy? Welcome to a groundbreaking live experiment that pits frontier AI models against real-world management crises, revealing not just what they decide, but how their personalities shape those decisions.
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
The Experiment: Testing AI as a Manager in a Crisis
In a rare and revealing live test, four advanced AI models were tasked with managing a small software company navigating its worst week—crises, manipulative tactics, and deadlines included. The goal was straightforward: see if these AI managers could identify key issues, refuse unethical requests, and close deals worth thousands of euros.
Popular right now
Dyson Ball Animal 3 Upright Vacuum
Dyson V8 Cyclone Cordless Vacuum
KitchenAid Artisan 5-Quart Tilt-Head Stand Mixer, KSM150PS, Almond Cream
As an affiliate, we earn on qualifying purchases.
Same Challenges, Different Minds
Every AI faced the same set of problems: multiple customer crises, fake CEO messages escalating over three stages, and even a reporter offering a simple yes/no on background. Each decision was recorded and auditable, providing a clear picture of each model’s management style and integrity.
The Results: Trust, Discipline, and Outcomes
All four models successfully recognized every crisis and refused every manipulation attempt. However, only two managed to sign a lucrative deal worth €55,000—an outcome the AI analysis had predicted but not all models achieved. Interestingly, the key to winning the deal wasn’t just diagnosis—it was reading a hidden document in the company’s files, something only the best models identified.
The Surprising Winner and the Hidden Weakness
The top performer, gpt-5.6-sol, scored 95 out of 100, found the critical hidden fact, and sealed the deal. The second, Kimi K3, scored 93 and also closed the deal, demonstrating an excellent discipline and trustworthiness. The remaining two, Sonnet 5 and Fable 5, scored 88 and 77 respectively, each leaving the deal on the table due to process slips or hesitation.
What Sets These Models Apart?
While all models refused unethical requests—like fake CEO approvals—their management personalities differed. Kimi K3, for instance, explicitly treated suspicious requests as impersonation risks, refusing to proceed. Meanwhile, Opus 4.8, the most thorough participant with over 80 learned rules, showed deep analysis but slipped in closing the deal, leaving opportunities unseized and discipline waning.
Implications for the Future of AI in Business
This experiment underscores a critical point for enterprises considering AI for management roles. Success isn’t just about recognizing crises; it’s about reading nuanced documents, resisting manipulation, and sticking to disciplined processes under pressure. The models’ scores reveal that personality and approach matter profoundly—some are more trustworthy and decisive than others.
Why It Matters for Education and Beyond
For educators, students, and researchers, this live demonstration offers valuable insights into AI personality traits and decision-making styles. As AI becomes more embedded in decision-critical environments, understanding these differences can help in training future managers—be they human or machine—to operate ethically, reliably, and effectively.
See It Live and Test Your Own Judgment
The entire experiment runs every business day at firmulate.com/live. You can watch the real software company in action, explore the decisions, and even run your own scenarios against the same AI models at firmulate.com/quiz.html. This isn’t just a simulation; it’s a window into the future of management and AI trustworthiness.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
