Opting out of using conversations for training is not the same as data isolation.
In Bezpieczna Kancelaria, the system controls whether an external model is needed and what material it receives.
The law firm's knowledge is kept separate from external models.
Artificial intelligence is incredibly convenient.
We open ChatGPT, drag a document into the window, type a question and receive a summary, analysis or draft pleading within seconds.
That simplicity may be its greatest risk.
It is easy to forget that a conversation window in a public AI system is not a private room at a law firm.
And this is not theoretical.
On August 24, 2026, Niebezpiecznik reported the story of Darren Shidy Zhou, who used ChatGPT while planning to kidnap, rape and murder his former partner. According to the report, the conversation was flagged by OpenAI's safety mechanisms and the information was shared with law enforcement.
We do not intend to criticize OpenAI for responding this way.
Quite the opposite.
This story shows something every lawyer using artificial intelligence should know.
ChatGPT is not a confessional.
What you enter into AI does not disappear when you press Enter
We have grown used to treating a chat like a conversation.
We type a question. We get an answer. We close the window.
Psychologically, it feels like talking to another person. Technically, it is not.
The content must be sent to the provider's infrastructure and processed according to its service rules and the safety mechanisms it uses.
OpenAI explicitly states that when its system detects users planning to harm others, their conversations may be referred for a specialized human review. If the team determines that there is an imminent threat of serious physical harm, the matter may be referred to law enforcement.
Moreover, even a feature called Temporary Chat does not mean that the conversation technically ceases to exist when the window is closed. OpenAI currently states that a copy of such a conversation may be retained for up to 30 days for safety purposes.
Files also have their own retention rules. OpenAI explains that in environments where the File Library is available, deleting a chat does not necessarily delete a file saved in the Library.
None of this means that OpenAI is acting improperly.
It means something else:
A lawyer should know exactly where they are sending information protected by professional secrecy.
For an ordinary user, this is an extreme situation. For a law firm - it is case material
A public AI safety system may encounter words such as murder, rape, kidnapping, weapon, threat, drugs and violence.
For the average user, this content may be highly unusual.
For a lawyer handling criminal, family or domestic violence cases, it may be a routine part of the case file.
A lawyer may have:
- a rape victim's testimony,
- a murder scene examination report,
- messages from a perpetrator threatening the victim,
- the victim's medical records,
- photos of injuries,
- an expert opinion,
- the defendant's statement,
- correspondence concerning violence against a child.
And they may want to use AI to analyze this material.
The issue is not that artificial intelligence should not have access to such information.
Quite the opposite.
AI can give lawyers the greatest advantage when working with case files that run to hundreds or thousands of pages.
The question is:
Where are we sending these documents, and who decides what an external model actually needs to see?
Opting out of training is not the same as data isolation
One of the biggest misconceptions about artificial intelligence is treating two different statements as equivalent:
“the provider does not use my data to train models”
and
“my data does not leave an environment I control”.
They are not the same.
We can opt out of using conversations to improve models. We can use a business solution. We can use an API. We can select specific privacy settings.
Each of these decisions can improve security.
But a fundamental question remains:
What exactly was sent to the external system?
If a lawyer uploaded a 400-page case file just to ask about a single procedural deadline, the problem arose before the model generated the first word of its response.
We reverse this relationship
In Bezpieczna Kancelaria, we do not want lawyers to have to ask themselves before every question:
- Can I paste this document into OpenAI?
- Does this model retain data?
- Did I enable the right setting?
- Is this mode private?
- Has the provider's policy changed since last month?
- Can I send the entire case file, or should I remove some of the data first?
That should not be the lawyer's responsibility.
That is why we are building Bezpieczna Kancelaria around the opposite architecture.
The law firm's data is kept in an isolated environment.
That is where information about clients, cases, documents and the law firm's work history is stored.
The lawyer does not have to copy this information into a public chat. They ask their question within Bezpieczna Kancelaria.
Only then does our system control whether an external model is needed for the specific task, which model can be used, what information it actually needs and what query will be sent to it.
This is a fundamental difference.
We do not hand the law firm over to the model. We give the model only what it needs to complete a specific task.
We formulate the prompt for the AI
The lawyer can ask:
“When is the deadline for filing a response?”
With public tools, the workflow often looks like this:
open ChatGPT -> upload a document or the entire case file -> ask a question.
In Bezpieczna Kancelaria, the lawyer's question first goes to the firm's environment.
The system knows the case. The system can find the information needed. And the system prepares a query that, if an external model is needed, will be sent on.
This means the external model does not need to receive the client's entire history just because the lawyer needs an answer to one specific question.
We control the boundary between the law firm and external AI.
Public AI chat
Users decide for themselves what to paste or upload. They need to know the service's current settings, understand its retention policies and remember to limit the scope of the data they share. Once the material is uploaded, how it is subsequently processed also depends on the provider's current policies.
Bezpieczna Kancelaria
The lawyer works within a single environment. The system controls whether an external model is needed and what information can be shared with it. The lawyer does not need to move the case file into a public chat window.
But what about a year from now?
This question is even more important.
Models change. Features change. Memory mechanisms change. Security systems change. Data retention rules change. Abuse detection methods change. Terms of service and laws change.
On August 19, 2026, OpenAI announced another safety processing solution for some customers using the API and Zero Data Retention mode. The company introduced Private Safety Processing, a mechanism designed to detect risk patterns across related interactions while preserving zero-retention commitments for eligible customers.
This is a good example of how quickly technology and its security mechanisms change.
What will change in six months?
What will be analyzed a year from now?
How will models work in 2027?
What new obligations will be imposed on their providers?
We do not know.
And lawyers should not be forced to keep track of it either.
Lawyers should not have to study OpenAI's algorithms
A lawyer should know the law. Know the case. Represent the client.
They should not also have to track each new version of the retention policies of OpenAI, Anthropic, Google or the next model provider that may offer the best solution on the market a year from now.
They should not have to know which mode retains data for a specific period.
They should not have to remember where to turn off history or training.
They should not have to check before every use whether the provider has just changed its security policies.
That is what the system is for.
If we switch AI providers tomorrow, the lawyer should still ask the question in exactly the same place.
If a model far better than today's becomes available a year from now, we should be able to use it without moving the entire law firm into its provider's environment.
If one provider changes its policies, we should be able to limit the information shared with it or stop using it.
That is the difference between using a public AI chat and having a dedicated AI environment for your law firm.
The biggest threat may be a simple "drag and drop"
It does not take a cyberattack.
No one needs to break into the law firm.
No one needs to steal a password.
A simple, everyday action may be enough:
dragging a case file into a public chat window.
From that point on, how this information is processed also depends on the architecture, policies and procedures of the entity we shared it with.
And law firm materials are a special category of information.
They contain data about health. Assets. Family. Sexuality. Disputes. Crimes. Litigation strategy. Trade secrets.
Sometimes they contain things the client has told no one but their lawyer.
That is why the question should not be:
"Is ChatGPT secure?"
That question is too simple.
The right question is:
"Should we share everything we know about our client with a public chatbot?"
Our answer is: there is no need.
An isolated environment is not a marketing add-on
You could create a law firm system and simply add a window that connects directly to a popular AI model.
It would be faster. It would be easier. It would be cheaper.
But it would not be Bezpieczna Kancelaria.
Because for us, artificial intelligence does not start with the question:
"Which model is the best?"
It starts with the question:
"How do we ensure the model gets exactly as much information as it should - and not a single piece more?"
That is why we separate the law firm's knowledge from external models.
The law firm keeps its data.
Our system controls the flow of information.
The system selects the information needed for the specific task and prepares the query sent to the external model.
And when the technology changes, we can change the model without changing the core principle.
The law firm's data remains on the right side of the boundary.
ChatGPT is a great tool. But it is not a law firm
The story reported by Niebezpiecznik is not a story about a bad ChatGPT.
It is about something much more important.
It shows that a public AI system provider has its own security mechanisms, procedures, terms of service and obligations.
And it can develop them.
It should develop them.
The problem begins when a law firm starts treating someone else's infrastructure as if it were its own closed environment.
It is not.
That is why the future of AI in law is not about banning lawyers from using the best models.
Quite the opposite.
We want to use the best models available.
OpenAI. Anthropic. Google. Models we do not even know about yet.
But we want to use them on our terms.
Without automatically handing over the entire law firm.
Without making security depend on whether a lawyer has just read the provider's latest terms of service.
Without expecting a lawyer to become an expert in artificial intelligence infrastructure.
The lawyer should ask the question. Bezpieczna Kancelaria should take care of what happens to the data next.
Because legal professional privilege should not depend on a checkbox in a chat window.
It should not depend on terms that may change tomorrow, either.
It should be built into the system architecture.
Sources
- Niebezpiecznik, „He planned to kill his ex-girlfriend, but ChatGPT alerted the police”, 24.08.2026: https://niebezpiecznik.pl/post/planowal-zabic-swoja-ex-dziewczyne-ale-chatgpt-poinformowal-o-tym-policje/
- OpenAI, „Helping people when they need it most”: https://openai.com/index/helping-people-when-they-need-it-most/
- OpenAI Help Center, „Temporary Chat FAQ”: https://help.openai.com/en/articles/8914046-temp
- OpenAI Help Center, „Chat and File Retention Policies in ChatGPT”: https://help.openai.com/en/articles/8983778
- OpenAI, „Offering Zero Data Retention for frontier models”, 19.08.2026: https://openai.com/index/offering-zero-data-retention-for-frontier-models/