Potomac University Hospital, Northern Virginia
October 12, 2026
Rosa Alvarez had already taken off her coat when the nurse told her the treatment would have to wait.
“How long?”
“We don’t know yet,” Leah Mercer said. “Dr. Morrow is checking something and is waiting for our clinical team to arrive. ”
Rosa looked at the empty hook beside her chair. Then she began putting on her coat to stay warm.
Across from her, her daughter closed a laptop.
“Is it her blood tests?”
“No. Those were reviewed.”
“Then what is it?”
Before Leah could answer, Elena Bieri appeared in the doorway.
She had come upstairs for a meeting about the hospital’s research software. Instead, she found three patients waiting, an unopened packet of IV tubing on a counter, and Leah looking relieved to see her.
“Are you the person we’re waiting for?” Rosa asked.
Elena stopped beside her chair.
“I’m here to help find out what happened.”
“Then I’ll leave my coat on.”
She smiled as though she were making things easier for Elena.
The research unit had six reclining chairs and a television tuned to a home-renovation program. On the screen, a couple argued about a kitchen island. Rosa’s daughter glanced at the time.
Leah led Elena into the corridor.
“The computer raised a warning overnight,” she said. “It found something in the records from earlier treatments. Owen wants it checked before we give any more.”
“What did it find?”
“It says three patients may have had a drop in their blood oxygen.”
“Are they all right?”
“The team is checking on them and reviewing their records. The staff assessed them when the readings happened. At the time, they thought the monitors were giving bad readings.”
Elena looked back through the doorway. Rosa was talking with her daughter and paused to briefly look down the corridor towards Elena.
“And now the computer thinks those readings might mean something.”
“Yes,” Leah said. “That’s why you’re here.”
Dr. Owen Morrow had printed the warning.
He laid it on the conference table between Elena and Samir Rao, the engineer who maintained the research software.
Possible safety concern: low oxygen readings recorded during three earlier treatment visits. Review recommended.
“The study drug might be affecting breathing,” Morrow said. “Or the machines might be giving us misleading numbers. I need to know what this warning is based on.”
Elena took the chair beside Samir.
Her doctorate was in computer science. She helped the hospital test its AI tools and investigate their mistakes. Decisions about patients belonged to Morrow and the clinical team.
The distinction mattered this morning.
“Show me what it found,” she said.
Samir opened three graphs on his laptop. Each line dipped and rose again.
“The assistant checks the study records every night,” he explained. “It looks for possible problems across visits—things that might be easy to miss when you’re looking at one patient at a time.”
“And last night it put these together,” Elena said.
Samir nodded. “Three similar dips. It treated them as a possible pattern.”
Morrow rubbed his eyes.
“One bad reading is something we see. Similar readings in three patients taking the same study drug deserve another look.”
The AI had helped before. Six months earlier, it had caught a laboratory result attached to the wrong patient’s record. A coordinator corrected the mistake before the research team used it.
Elena remembered Morrow’s pleasure at that discovery. Software doing something useful, quietly, without anyone needing to be impressed by it.
This morning, its warning had stopped the unit’s work.
“Can you tell me if the comparison is sound?” he asked.
“I’ll start with the original records.”
“How soon?”
“Give us an hour for a first answer.”
He picked up the warning.
“I’ll talk to the patients.”
In an empty examination room, Samir placed his laptop on the counter.
Elena pulled up a chair.
“Take them one at a time.”
The readings came from the small clips placed on patients’ fingers to estimate blood oxygen. They were useful, but a loose clip or a moving hand could interfere with the measurement.
The first patient had reached for her phone.
Beside the sudden dip, the monitor had recorded a warning of its own:
Poor signal. Reading may be unreliable.
The nurse’s notes described checking the patient and adjusting the clip.
“Now show me what the AI received,” Elena said.
Samir opened a second window.
The low number was there.
The warning about the poor signal was not.
Elena leaned closer.
“That’s the problem.”
Samir switched between the two windows.
A piece of software prepared the monitor records before the AI reviewed them. It had copied the oxygen readings but left out the monitor’s warnings.
“The AI got the number,” Elena said, “but lost the reason not to trust it.”
Samir nodded slowly.
“So it read this as a drop in oxygen.”
“Instead of a measurement the monitor itself had questioned.”
They checked the second patient.
The same thing had happened. The monitor had warned of a poor signal. That warning had disappeared before the record reached the AI.
Elena felt her shoulders loosen.
“Two of the three.”
Samir opened the last file.
This time, there was no poor-signal warning.
The nurse had assessed the patient, and a separate measurement had been reassuring. But nothing in the record clearly explained the brief dip.
Elena read the notes again.
“This one stays open.”
“Even with the mistake in the others?”
“The mistake explains those two. It doesn’t explain this.”
They restored the missing warnings and ran the review again.
The AI withdrew its claim of a pattern across three patients.
One unexplained reading remained.
A residual, Elena thought. The part left over after they had accounted for the rest.
She called Morrow.
“We found an error in how the software prepared the records. Two unreliable readings were presented to the AI without their warning labels.”
“So the three-patient pattern doesn’t hold up.”
“Correct. But we still have the third reading to check.”
“All right,” he said. “Send me exactly what you found.”
After she ended the call, Samir scrolled down.
“There’s something else.”
Beneath the revised result was a reference to an equipment test.
The AI had found a similar dip in a monitor being tested with a machine that imitated a human pulse.
No patient had been attached.
Elena read the reference twice.
“If a monitor can produce this during a test,” she said, “the equipment could be causing the dip.”
“It could.”
It was a useful lead. It might help explain the remaining reading.
Then Elena noticed where the test had been published.
“This isn’t in our study records.”
“No.”
“Where did the assistant get it?”
Samir opened the history of its overnight work.
“I’ll find out.”
Rosa’s daughter was standing beside the coffee machine when Elena returned upstairs.
“Any news?”
“We found a mistake in the computer’s comparison. Two of the readings weren’t reliable, and it hadn’t been given that information.”
“So it was a false alarm?”
“Part of it was. There’s still one reading the team needs to understand. Dr. Morrow will explain what that means for today.”
The woman looked down at her phone. A meeting reminder filled the screen.
“I told work I’d be back by eleven.”
“I’m sorry.”
“I know it isn’t your fault.”
Elena wasn’t ready to agree with her.
Inside the treatment area, Leah had brought Rosa a blanket. Rosa was smoothing it over her knees with the heels of her hands.
A paper cup sat just beyond her reach.
“Would you mind?” she asked.
Elena lifted it and waited while Rosa settled her fingers around it.
“I used to make wedding cakes,” Rosa said. “Now a cup of coffee requires planning.”
“Wedding cakes?”
“Mostly. Birthdays too. People always wanted the same cake they’d seen in a picture, only for half the price.”
Elena smiled.
“My granddaughter’s getting married in May.” Rosa looked at her hands. “She says a bakery is fine.”
Elena waited.
“She means it kindly.”
Morrow came in and pulled up a chair.
He explained what they had found. The computer’s warning had overstated the evidence, but the team still had something to check. He had asked the study’s safety reviewers and the hospital’s equipment specialists to examine it.
Rosa would not receive her treatment that morning.
“Am I out of the study?”
“No,” he said. “This delay doesn’t mean you’re out.”
“Did I do something wrong last time?”
“No. You didn’t.”
He went through her care plan and told her who would call with an update. Her daughter made a note on her phone.
Rosa listened, then turned to Elena.
“Was the computer right or wrong?”
“It made a mistake about two of the readings. But it also brought another one to our attention.”
Rosa thought about that.
“So it was right to ask.”
“Yes,” Elena said. “It was right to ask.”
By lunchtime, Samir had traced the equipment reference.
Elena found him in her office with a packet of crackers beside his keyboard.
“Our assistant used the outside research service,” he said.
“Pelorus?”
He nodded.
The hospital paid Pelorus to find scientific papers, equipment manuals, and other reference material. The AI assistant had permission to use it.
That much was expected.
Samir turned the screen toward her.
“Last night, our assistant asked whether this type of monitor had ever produced a sudden dip during equipment testing.”
“And Pelorus found the test.”
“Yes. But it also had another AI compare the recordings and write an explanation.”
Elena sat down.
“Our assistant asked another AI to help investigate.”
“Through Pelorus, yes.”
“Does the report tell the reader that?”
“No. It gives the result and the reference.”
She looked at the short paragraph the hospital had received. It was neatly written and sounded assured.
Behind it were decisions someone on the team needed to check.
“Was any patient information sent out?”
“I’ve blocked new outside requests while we check. The request I’ve traced asked about equipment behavior. It didn’t contain names or patient records.”
“Check the rest too.”
“I will.”
Samir brought up the service description. It called the programs doing this work agents.
Elena pointed to the word.
“This is the part we need to explain better when we train people.”
“What an agent does?”
“Yes. People hear ‘AI assistant’ and imagine something that answers a question. This one can take steps to complete a job. Search records. Use another service. Ask another program for help.”
“Within the permissions it’s given.”
“And we need to know what those permissions allow.”
She looked again at the equipment comparison.
The outside program might have done excellent work. Its finding might help the doctors. But nobody at the hospital had yet checked how it had chosen the test or decided the readings were similar.
“Get the full record from Pelorus,” she said. “What it searched, what it found, and how it reached this answer.”
Samir began typing.
Elena opened her incident report.
She wrote the first problem plainly: the software had dropped warnings that should have accompanied the monitor readings.
The second took longer.
Finally, she typed:
The report includes analysis performed by an outside AI. The hospital received its conclusion without enough detail to check its work.
She read the sentence once more.
That was the issue. They had to be able to follow the work before they relied on the answer.
At three, Leah arrived with two sandwiches.
“You can choose,” she said. “But you have to eat one.”
Elena moved a stack of papers.
“Did Rosa get home?”
“Yes. Her daughter sent a message.”
“Was she angry?”
“Disappointed.”
Elena unwrapped a sandwich.
“I keep thinking about her hands.”
“She was hoping this treatment would help.”
“We should have caught the missing warnings before the software went into use.”
“Yes,” Leah said.
Elena looked up.
Leah took the other chair.
“What are you doing about it?”
“Samir’s fixing the problem. We’re checking the earlier reviews to see whether it happened anywhere else. And we’re investigating how the assistant got that outside answer.”
“All right.”
For a while, they ate without speaking.
Then Leah nodded toward Elena’s report.
“Put the patients in there.”
“They’re in the study records.”
“I mean today. The delay. Rosa’s daughter missing work. The man who arranged a ride and then had to call for another one.”
Elena looked at what she had written.
Most of the page concerned software.
She added the delayed treatments and the patients who had gone home.
At five eighteen, Samir returned.
Elena had put on her coat. She took it off again when she saw his face.
“Pelorus sent the activity record.”
“What did you find?”
“The outside AI didn’t stop with the monitor test.”
He set the laptop on her desk.
After finding the test, the program had searched for similar interruptions in other kinds of equipment. Most of the results had gone nowhere.
One had interested it enough to keep looking.
An instrument used in ocean research.
Elena looked up.
“What does that have to do with a hospital monitor?”
“We don’t know.”
He opened the record.
The ocean instrument used light to examine water samples. Its recording contained a brief dip that the outside AI had marked for comparison.
“Could they use the same component?” Elena asked.
“Possibly. Or similar software. It could also be a resemblance that means nothing.”
She studied the two graphs.
They looked alike in places. That was a reason to check, not an answer.
Then Samir moved to the next entry.
Further analysis requested.
Elena pointed at it.
“Requested by whom?”
“The outside AI.”
“From whom?”
“That’s what I haven’t been able to establish.”
She sat back.
The sequence was clear now.
The hospital’s assistant had asked Pelorus about a monitor. Pelorus had used another AI to investigate. That AI had found something in an ocean instrument—and passed a new question onward.
“Did anyone here approve that last request?”
“Not individually. It may be covered by the service’s general permissions. I’m checking.”
“When you blocked our assistant’s outside requests, did that stop this work too?”
“I don’t know. This request had already been sent.”
Outside her office, a cart rattled down the corridor. Someone laughed near the elevators. The hospital was changing shifts.
Elena looked at the paragraph they had received that morning.
It gave no indication of how far the investigation had gone.
She thought of Rosa, waiting with her coat off.
The first question had been simple: was there a reason to delay her treatment?
By morning, one program had asked another for help. That program had searched beyond medical equipment and sent a question somewhere else.
It might all have an ordinary explanation.
They still needed to know where it ended.
“Save the whole exchange,” Elena said.
Samir nodded.
“Then find out where that last request went.”
Join the conversation
No technical background is needed. Choose a question that interests you and share your thoughts in the comments—or use these questions with your own reading group.
- Whose situation stayed with you most: Rosa’s, her daughter’s, Leah’s, Dr. Morrow’s, or Elena’s? Why?
- Was Dr. Morrow right to pause the infusions? What information would you have wanted before deciding?
- Rosa says the computer was “right to ask.” Can a warning be partly wrong and still be useful?
- When one AI system asks another for help, who should be responsible for checking the answer?
- What would you want Elena to investigate next?
Please keep this discussion to Chapter One so new readers can join without encountering later spoilers. Different viewpoints are welcome; challenge ideas with curiosity and treat one another with respect.
A note about the story
Potomac University Hospital, its patients, the clinical trial, and the events described here are fictional. The story explores questions about AI assistance, human responsibility, and how people make decisions when the evidence is incomplete.
Future discoveries in the novel are possibilities imagined for this story, not predictions.
Leave a Reply