The smell of wet wool and radiator dust filled the small office, a scent that always seemed to intensify when the radiators hissed at four in the afternoon. Sarah pulled her coat closer around her shoulders, the dampness from the walk back from the courthouse still clinging to the fabric.

She was holding a physical file, the edges frayed and the manila surface stained with a ring from a coffee cup that wasn’t hers. Inside this file was an email, printed out on cheap paper that felt thin and overly smooth. It was from a woman named Elena, written back in February, and it contained three paragraphs of devastatingly honest prose regarding her husband’s cognitive decline and the specific date he had forgotten their daughter’s name.

Sarah had been tasked with summarizing this case for a Monday morning briefing. The summary needed to be tight, clinical, and persuasive. She looked at the laptop sitting on her desk, its lid closed for the moment, and then back at the printed email. She knew that if she opened that lid and fed Elena’s words into the tools she usually used, the summary would be finished in under ninety seconds.

The “efficiency” would be undeniable. But as the radiator clanked and the smell of hot metal rose to meet the damp wool, Sarah found herself stuck on a thought she usually managed to ignore: Elena didn’t know about the laptop.

Elena had entrusted her story to Sarah because Sarah was a lawyer. She had given her consent for Sarah to use this information to win a case. She had not, however, given her consent for her husband’s most vulnerable moment to be processed by a cluster of servers in a desert she would never visit, owned by a corporation that had never met her. Sarah was about to make a disclosure decision on behalf of a person who was not in the room, and there was no box to check that would ask Elena if she was okay with her family’s tragedy becoming a training weight for an algorithm.

The central lie of the modern interface

We frame the entire debate around the “user.” We talk about user privacy, user consent, and the user’s right to delete their data. But in any profession that involves the care of others-law, medicine, insurance, social work-the person sitting at the keyboard is rarely the person whose life is actually at stake in the data.

The “user” is merely the pipe. The content belongs to someone else, yet the architecture of the software treats the professional as the sole sovereign of the information. It is a convenient arrangement for everyone except the person the information describes.

Ian Y., an insurance fraud investigator I know, once told me that he tried to meditate for twenty minutes every morning to clear his head of the “filth” he sees in his line of work. He usually gives up after eight minutes because he keeps checking his watch to see how much “peace” he has left to earn.

“We’re taking out loans on people’s lives without their signatures.”

– Ian Y., Fraud Investigator

Ian views data differently than most. To him, a file isn’t a collection of bytes; it’s a collection of liabilities. He told me recently that the industry is currently operating on a massive “privacy debt.”

“In a room of 14 people, 11 of them have their most intimate secrets currently sitting on a server they didn’t know existed, simply because their consultant or their agent liked a new productivity plugin,” he said, tapping a thick stack of medical records.

This is the reality of the shadow data trail. For every one professional who clicks “I Agree” on a set of Terms and Conditions, there are dozens of third parties whose consent is being assumed by proxy.

The problem is structural. Most software is designed for the individual consumer. If I want to upload my own grocery list to a cloud-based AI, that’s my prerogative. I am the user, and I am the subject. But when Sarah uploads Elena’s email, the subject and the user are decoupled. The consent framework remains a one-to-one relationship between the software provider and the account holder, completely bypassing the actual owner of the secret.

The Custodian’s Burden

Entrustment is the transfer of a burden without the transfer of ownership. Therefore, if a professional is entrusted with a secret, they are the temporary custodian of a weight that does not belong to them, which means that any action taken to lighten that weight-such as using an AI to summarize a difficult file-must be weighed against the original owner’s expectation of silence.

If we define privacy as the right to control the boundaries of one’s own story, then the moment a professional moves that story into an unencrypted, third-party environment, they have redefined the boundary without the author’s permission.

We tell ourselves that because the data is “de-identified” or because the AI “doesn’t remember,” it’s fine. But Ian Y. sees the cracks in that logic. He knows that in the world of insurance fraud, “anonymity” is often just a lack of effort.

With enough data points-a diagnosis, a date, a zip code-the “anonymized” person can be reconstructed with 92% accuracy. The professional is making a gamble with someone else’s chips, betting that the convenience of the tool is worth the erosion of the client’s wall of confidentiality.

Sarah looked at the clock. It was 4:18 PM. The deadline was Monday, and she had four other summaries to write. The pressure to be productive is a physical weight, much like the heavy winter coat still damp on her shoulders.

The industry demands speed, but the ethics of her trade demand a slowness that feels like failure in a digital age. This is the trap: we have built tools that are incompatible with the fundamental duty of the professional. To use them is to betray the client; to ignore them is to fall behind.

There is a way out, of course, but it requires a fundamental shift in how we think about the gateway to these technologies. It requires moving away from the “user as owner” model and toward a “custodian” model. This is where tools that prioritize local encryption and total anonymity for the source material become the only ethical choice for someone in Sarah’s position.

If Sarah were using an Encrypted chat gpt service, the dynamic changes. The data wouldn’t be “given” to the provider; it would be processed in a vacuum, encrypted before it even leaves the room, and stripped of the digital fingerprints that could lead back to Elena or her husband.

The Ghosts in the Machine

But most people aren’t using those tools yet. They are using the ones that come as the default, the ones that treat data as a raw material to be harvested. We are currently living through a period where the party with the strongest interest in the question-the client-is the only party with no way to raise it.

They aren’t in the room when the software is chosen. They aren’t in the room when the “Upload” button is clicked. They are the “absent third party,” whose lives are being processed for the sake of a shorter Friday afternoon.

Ian Y. told me that he once caught a guy in a disability fraud case because the guy’s physical therapist had uploaded a “success story” to a public-facing AI to help write a marketing blog post. The therapist didn’t name the patient, but they described the injury and the recovery timeline so specifically that Ian was able to match it to his claimant in under five minutes.

The patient never knew his recovery had been used as “inspiration” for a blog, and he certainly didn’t know it was the reason his claim was eventually denied. The therapist thought they were just being efficient.

This is the hidden cost of the “convenient” consent. It’s not just about data leaks in the traditional sense; it’s about the loss of context and the loss of agency. When we feed someone’s life into a machine, we are stripping away their right to be forgotten. We are making them part of a permanent, searchable record, even if we tell ourselves we’re not.

Sarah finally opened her laptop. The screen glowed, but she didn’t open the browser. Instead, she stared at the email from Elena. She thought about the daughter who was mentioned-the one whose name the father had forgotten.

That name was a piece of information that belonged to that family, and no one else. If Sarah used the standard tools, that name, or the shadow of it, would exist in a training set forever. It would be a tiny, insignificant bit of weight in a multi-billion dollar model, but it would be there, displaced from its home.

She began to type, but she did it the old-fashioned way.

She summarized the case herself, sentence by sentence, feeling the friction of the work. It was slow. It was tedious. Her back began to ache, and the radiator finally stopped its hissing, leaving the office in a sudden, heavy silence. She knew she wouldn’t be home in time for dinner. She knew her colleagues would finish their briefings hours before she did.

Honoring the Silence

But as she worked, she felt a strange sense of relief. By choosing the slow path, she was honoring the only person who couldn’t voice their objection. She was closing the “Consent Gap” by acting as a true custodian rather than just a user.

The problem with modern professional work is that we have been trained to value the tool over the trust. We have been told that as long as we click “Agree,” we have done our due diligence.

The reality is that “Agree” is a lie we tell ourselves to justify our own convenience. Until the design of our tools recognizes the standing of the absent third party, the entire architecture of digital permission is addressed to the wrong person.

We are making decisions for people who have no idea we are even choosing, and in the quiet of a damp office at 5:30 PM, that realization is enough to make any professional pause. The deadline is still Monday, but the secret is still a secret. And in Ian Y.’s world, or Sarah’s, or yours, that is the only thing that actually matters at the end of the day.