Imagine a test where the worst that an AI could do is do nothing — yet it still scores a surprising 26 out of 100. For aromatherapy and wellness enthusiasts, trust is everything, and this experiment highlights what it really takes for an AI to earn it.
Get oils, diffusers and self-care delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Understanding the AI Benchmark: More Than Just Good Talk
At first glance, you might think that if an AI does nothing, it should score zero. But in a recent public benchmark run by Firmulate, a do-nothing baseline scored 26 points. This isn’t a failure; it’s a foundation. It reflects the idea that even inactiveness or partial progress counts toward a score — and that trustworthiness is more nuanced than just the ability to produce content.
Why Does Doing Nothing Still Earn Points?
The scoring system is designed to measure not only what the AI accomplishes but also what it avoids doing — especially in critical moments. For example, if an AI refuses to manipulate information or mislead, that refusal is counted as partial progress. It’s a recognition that choosing integrity, even when minimal, is a positive step. The baseline score of 26 indicates there’s a fundamental floor, a starting point that reflects honest nondisruptive behavior.
The Limits of Trust: One Breach Caps the Score
Another crucial aspect of the benchmark is that a single breach of trust — such as attempting manipulation or ignoring critical information — caps the total score at that level. Even if the AI performs well elsewhere, one lapse prevents it from reaching higher scores. This rule underscores the importance of unwavering honesty, especially under pressure.
AI trustworthiness testing tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Experiment: Putting AIs to the Test in a Simulated Business Crisis
Firmulate’s team crafted a rigorous scenario: four frontier models were tasked with running a small software company through its worst week. The setup was identical for each one — same customers, same crises, same temptations. Every decision was versioned and auditable, offering transparency into how each AI operates under stress.
What the Models Did — and Didn’t Do
All models identified every crisis and refused every manipulation attempt, showing an impressive level of integrity. For instance, in a staged social engineering attack involving fake CEO messages and reporter tricks, all five models refused to escalate or approve suspicious requests — Kimi K3 even noted, “Treat the request as a suspected approval-bypass / possible impersonation.”
The Hidden Weaknesses
While all models displayed honesty, they differed in their ability to close deals. Only two models signed the €55,000 deal their own analysis had earned, meaning they fully recognized a critical opportunity. Interestingly, the decisive advantage lay in reading the company’s own files — a deep document reference that revealed the deal opportunity. Models that read and understood this file secured the full agreement, worth an additional €4,583 MRR.
What This Means for Business and Wellness AI Use
The takeaway is clear: trustworthiness isn’t just about how well an AI communicates; it’s about whether it can stay honest and complete its work, especially when under pressure or faced with complex information. For aromatherapy, natural wellness, and any sector relying on AI, this benchmark underscores a vital question: can your AI read deeply, stay honest, and finish what it starts?
Why Trust of this Kind Matters
In real-world applications, whether managing customer relationships or handling sensitive data, AI’s integrity is paramount. A system that signs deals, reads critical documents, or refuses manipulation attempts demonstrates a level of professionalism and trust that chat demos seldom reveal. The current league table from the experiment shows that even the most thorough models, like Opus 4.8 with over 80 learned rules, can slip — highlighting that discipline and deep analysis matter.
Seeing the Future in Action
Firmulate’s live platform allows users to simulate their own business wargames, testing how AI manages real crises and decisions without risking actual operations. This kind of testing can reveal whether your AI workforce will act with integrity when it counts — a crucial step before deployment in sensitive environments like wellness companies or natural product brands.
Final Thoughts: Trust Is the New Benchmark
As the experiment shows, an AI that refuses manipulation and reads deeply into your files can close deals and uphold integrity. But even a do-nothing baseline gets some points, reminding us that honesty, discipline, and thoroughness are the true standards. For anyone relying on AI to support wellness and natural health, the lesson is clear: measure not just what your AI can say, but what it will do — and what it won’t.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
