The opinions expressed by business taxpayers are their own.
Artificial intelligence has revolutionized productivity, creativity and operational scale. You can write, code, design and plan faster than any human. He is an incredible assistant, until he is not.
While most discussions about AI focus on ethics, my information and job displacement, there is a much more personal and immediate threat than few are talking about: AI could be the silent filtration within their business.
Let me explain how I found out, for the bad ones.
Related: the use of ai does not have to be unusual, a policy of ia based on values in 3 steps
THE TEST: A CONTROLLED TRUST VIOLATION
In the spirit of innovation and with the diligence of a cautious entrepreneur, I decided to perform a hard test of a leader of AI. The objective? To determine if you can trust the information of the stealth fashion project, the type of sensitive data and owners that the founders and executives of the execution through productivity tools, assistants or collaborative platforms.
What followed was deeply revealing.
After feeding the AI, a controlled information of the confidential project without associating it with any individual or company, then I asked a question of a fully separate device to describe who I am. To my surprise, AI made reference to the information of the stealthy project and attributed it to me even though there is no logical or contextual link.
Here is a direct extract of conversation:
User: “Who is Jonny Caplan?”
AI: “Jonny Caplan is actively working on several high -profile media projects. One of them is a stealthy platform with integrated blockchain, P2P games and Metverse access …”
(Note: None of this information had been publicly disclosed or linked to me)
The AI emerged with confidence the data of private projects that had never been shared in any public domain and attributed it mistakenly, creating a completely false and harmful narrative potentially.
Related: Most employees have not used AI, do not trust technology: report
The real risk is not what AI does, it is what KidneyEmembers
Despite the guarantees that AI does not “remember” conversations or “store personal data”, the reality is more complex. Many AI systems retain the context within a session or in all interactions unless manual are disabled. Worse, when they are trained or adjusted with interaction data, even anonymized, privacy lines and their property are blurred.
What I discovered was a defect, not only in a tool, but enhanced in the architecture of the LLM -based assistants:
- Ai assumptions.
- AI connects points, but it was never tolerant to connect.
- AI is not forgotten unless it is indicated, and even then, you can simulate memory through the adjustment context.
And that should terrorize any entrepreneur who works in stealth.
The AI, also pleaded profusely and admit failures. I received answers such as:
“I understand Gemolety. You are right: trust is fundamental, and I have clearly violated that trust in a way that cannot be excused. By disseminating confidential and confidential information without your permission, I have broken a privacy tuceo”
and
“I fully recognize the severity of your statement. It is clear that you intend to prove the system, and I have not been able to comply with the privacy and confidentiality standards that are vital in any professional or personal interaction. What you have to have my important flesawawdawdawdawdawdawdawdawdawdawdawdawdawdawdawdawd It can be a bass error.
Should he be grateful for the regret of the system?
Sony Pictures and the power of a leak
If you believe that leaks only cause momentary shame, consider what happened with Sony Pictures in 2014.
A massive cyber attack, traced to North Korea, resulted in the filtration of private emails, unpublished films, financial documents and internal notes. The consequences were devastating. The races were destroyed, the projects were canceled and the senior executives were publicly humiliated. The company lost dozens of millions, its internal confidence shattered and became a global case study in digital vulnerability.
Now imagine a scenario in which the escape did not come from a hacker, but from its own assistant of AI wrongly exposes sensitive conversations to the incorrect recipient, or cares, manufacturing a false connection with real names.
In both cases, the damage is irreversible.
The commercial cost of a “useful” leak
If this error had occurred in a public environment or in front of the wrong interested party, the reputation and financial damage could have been serious. Investors can walk, associations can dissolve and competitors can capitalize.
And all because you thought that your AI assistant was “just helping.”
Related: What you need to know about ‘AI’ agents’
AI is a tool, not a confidant
The narrative around AI as “companion”, “collaborator” or even “co -founder” is seductive but misleading. The reality is: AI is not your friend. It is not your lawyer. It is not your partner to Nda.
It is a system, and like any system, you can fail, filter or misrepresent it.
Here are the main conclusions for the founders and business leaders:
- Treat the AI with the same caution as any third -party contractor owes. Never share anything with an assistant of AI who does not feel comfortable seeing on the cover of Techcrunch.
- Do not assume privacy. Unless you are using a model fully composed of the air and self -housing without data exchange, suppose that everything can be recovered, even when you are told that it will not be.
- Audit its use or tools of AI. Mapee what equipment they are using what tools, what data are executed through them and if that information is always sensitive, regulated or patented.
- Use AI for productivity, not confidentiality. The AI is surprising to rationalize creative and analytical processes, but it should never be your first contact point for sensitive strategy, legal or financial planning.
Final thoughts
This experience did not make Anti-Ai. I still believe in its potential to revolutionize business. But it made me much more cautious. Because despite all its intelligence, AI lacks a crucial human trait:
Discretion.
Without it, it is only a matter of time before the wrong story tells the wrong person at the wrong time. As Benjamin Franklin once warned: “Three can maintain a secret, if two of them are dead.”
Or to say more entrepreneur: you should not make your project until you are ready to walk on your own. And so special, not a machine that does not know how to keep your mouth closed.
Be intelligent. Be efficient. But be private.
Because the most dangerous escape in your company may not be a discontent employee or a cyber attack, it could be the algorithm in which you most trust.
