Present day. Training centre near KLCC, Kuala Lumpur.
The room was bright in the way that training rooms always are — large windows, arranged tables, name cards, enough cable clips to suggest someone had thought carefully about the power points. Through the glass, the KL skyline held its usual Wednesday-morning posture: the Towers in the distance, the LRT threading between buildings, the city going about its business below.
Maya Chin had taken the second row. Not the front — she wasn't that kind of person — but close enough that she wouldn't miss anything. Laptop open, notebook beside it, a line already drawn down the centre of the first page. One column for things to remember, one for things to act on. Both still empty.
Maya was thirty-one and had been at BrightBridge for four years. BrightBridge Consulting was a management consulting and training firm in Petaling Jaya — ninety people, RM 20 million in annual revenue, a reputation built over thirty years on the kind of precision that made government ministries and GLCs renew their contracts without being asked twice. The firm had been founded in the mid-nineties by Dr. Krishnamurthy, whose name still carried enough weight in Malaysian professional services that people who had worked with him back then mentioned it with a particular quietness. The firm now belonged to his daughter, who had inherited it and carried it forward. Dr. Rohini had been running BrightBridge for long enough that most people no longer thought to separate the two.
Maya designed and delivered training programmes — TNA reports, programme proposals, post-training evaluations, her name on every deliverable. She was good at her job and knew it. That, as it turned out, had been part of the problem.
Amir was somewhere in the back half of the room. She knew he was there but hadn't looked. They had arrived separately and there was an unspoken agreement between them that this was not a team exercise. They were here to learn, individually, what they had not known they were missing.
The trainer was younger than most of the room expected — early forties, the kind of person who had clearly spent years doing the work before standing in front of it. He had a marker in one hand and had not yet uncapped it. He faced the room and it settled.
"Before we look at what AI can do for you — I want to ask something more uncomfortable. What has AI already done in your name — that you didn't check?"
The room absorbed this in the way of professionals caught off guard who will not show it. Someone stopped moving their mouse. Maya looked at her empty columns.
In your name.
The phrase did its work quietly. And sitting in the second row with the KL skyline behind her, Maya let herself remember.
Three weeks before the training. BrightBridge office, Petaling Jaya. A Wednesday.
The MoH brief was on one screen. A blank document on the other. Training Needs Analysis for the Ministry of Health — digital skills gaps across administrative and clinical support staff. Maya had done TNA work before, many times. She knew the structure. What she didn't have was time.
Three working days. Twelve pages. 800 staff across three departments. A client whose legal team cross-referenced everything.
She stared at the blank document for forty seconds. Then she opened ChatGPT.
The output arrived fast. Confident, formatted, exactly the shape of a document she would have written herself given twice the time. It had statistics, citations, the right sector vocabulary. It read like someone who knew the brief.
She copied the executive summary into her document, adjusted the header, and moved on.
She did not verify a single figure. She did not know she was supposed to.
The same Wednesday. BrightBridge office, late afternoon.
By the time the floor had thinned out, only the lamp on Amir's side of the office was still on.
He was fifty-two and had been at BrightBridge for nine years, with thirteen years in government-linked consulting before that — long enough that reading a client deliverable had become a physical habit. Slowly, from the beginning, without skipping, the way you read something when your name will be the last line of defence. He had built BrightBridge's public sector division from nothing. Four government ministry accounts. RM 6.5 million in annual revenue. The MoH relationship alone — four years, multiple contracts, the kind of trust that arrives slowly and departs quickly.
He read the executive summary Maya had submitted. Then he read it again.
"67% of Malaysian public sector employees report insufficient digital training (MoH Benchmarking Study, 2023)."
He leaned back.
He had worked with MoH for four years. He received their newsletters. He attended their briefings. He did not recognise this study.
It was possible he had missed it. He kept that possibility separate from what his instincts were telling him, the way careful men do. He did not send the report. He created a folder on his desktop, named it HOLD, and dragged the file in. He typed Maya a message: "Can we talk tomorrow? Need to check a few things." He did not say which things.
He drove home to PJ in the rain.
That Wednesday evening. Amir's home, Petaling Jaya.
His daughter was still at the table when he got in, laptop open, a mug of tea going cold at her elbow.
Putri was in her twenties — she had her father's stillness and her mother's directness, which made her, in Amir's considered opinion, better equipped for most situations than either of them had been at her age. She worked as a Business Analyst at a fintech company in KL, a forty-minute LRT commute she made because she said driving into the city was a choice only people with too much optimism would make.
She looked up and immediately looked at him more carefully.
"What happened?"
"Nothing happened."
Which was accurate. He sat across from her and opened his tablet.
"A report. One of my consultants. Some statistics I don't recognise."
Putri put down her fork — not her phone, not her pen, her fork — and looked at him across the table.
"Did she use AI to write it?"
He looked up. The question was so direct and so precisely aimed that for a moment he just sat with it. He hadn't asked Maya. It hadn't occurred to him to ask — he had assumed his team knew what they were doing with AI tools, in the general way that managers assume things they haven't individually examined. The assumption looked different sitting between them at the dinner table.
"I don't know."
Putri opened her laptop and reached into her bag for a folded handout. The month before, she had attended GenAI for Practitioners — a two-part AI training programme run by P2P Talent Development near KLCC. She had enrolled on her own time because the description matched something her fintech team needed and she had wanted to see it for herself.
She smoothed the handout on the table.
"Let me show you what they taught us."
That Wednesday night. The dining table.
Putri worked through it without drama, the way she worked through everything.
Traffic light classification first — the report contained specific citations from government sources, which made it Red: human verification required before submission. Then the forensic read: she found patterns in the first three pages she recognised from the training lab. Confident passive voice. Citations without page numbers. Statistics presented as findings rather than estimates. Recommendations that addressed the sector rather than the specific brief.
She extracted eleven claims and flagged four as high priority. Then she verified each one.
The MoH Benchmarking Study, 2023: no results across three search variations. Ministry of Health publications portal, national health data repository, full Boolean search. Nothing. The study did not exist.
The WHO Digital Health Framework citation: the document was legitimate. The specific figure attributed to it was not in the document. The AI had cited a real source for a number that wasn't there.
A workforce readiness statistic attributed to a 2022 industry survey: she found the survey. The published figure was 34%. The TNA stated 61%. The number had not been fabricated — it had been inflated, plausibly, in a way that would survive a casual search but not a careful one.
The subtlest finding. The recommendations were calibrated to a generalised Malaysian government ministry, not to a health ministry with clinical workflows and patient data considerations. The report had been written for a client that didn't quite exist.
Putri closed her laptop.
"Three of the four key statistics are wrong or unverifiable. One cited source contained a different number entirely. The recommendations don't cover the people the client specifically asked about. And the whole thing was written for a generic government ministry rather than a health organisation with clinical staff and patient data."
"The AI didn't invent all of it. Some of it was real information used inaccurately. That's actually harder to catch than a complete fabrication."
Amir was very still.
This was not a routine report. This was the TNA for MoH's Digital Transformation Unit — approximately 800 staff across three departments, a brief built from three rounds of stakeholder interviews that he had personally arranged. The contract was worth RM 95,000. Behind it, contingent on the quality of this deliverable, was a follow-on training design contract worth RM 220,000 more. It was the largest new account his division had won all year.
He thought about the MoH director — a careful, methodical man who had given BrightBridge three rounds of stakeholder interviews on the basis of a relationship Amir had spent four years building, one meeting at a time, one delivered commitment at a time. He thought about 800 staff across three departments whose real training needs were now sitting in a document that cited a study that did not exist, inflated a survey figure by 27 percentage points, and had quietly erased an entire category of clinical administrative workers from its recommendations.
He thought about what happened to a four-year relationship when trust broke at the point of delivery.
"If I had sent this to MoH—"
Putri looked at him steadily. She did not finish the sentence for him.
"Ayah — the AI didn't lie on purpose. It doesn't know how to say it's not sure. It produces what a correct answer looks like. The submission without checking — that part can be fixed."
She slid the handout across the table. He picked it up and read it slowly.
The next morning. BrightBridge office, meeting room.
The meeting room had a glass wall and half-drawn blinds. Amir arrived first and set the printed report on the table. Maya came in and read the room the way she read all rooms — quickly, without showing it — and sat across from him with her hands flat on the table.
He walked through it. Calmly, specifically, one item at a time. He showed her the screen. He did not editorialise.
Maya did not interrupt. When he finished she looked at the report rather than at him.
"I didn't know you had to verify the statistics. I thought if the AI cited something, it had looked it up somewhere."
"That's what it's designed to look like. It's not what it's doing."
"What about the data I uploaded?"
He told her. The free tool. The staff names and role categories she had uploaded to give the AI context. What it meant, legally, when data left BrightBridge's systems and was processed on servers outside Malaysia. The PDPA. The exposure she hadn't known existed.
She went very still. Not dramatic — Maya didn't do dramatic. The stillness of someone recalculating.
They corrected the report together over the next forty minutes. He searched. She rewrote. The unverifiable statistics came out. Verified figures, sourced and footnoted, went in. The frontline clinical administrative staff reappeared in the recommendations. The generic ministry framing was replaced with language that reflected an organisation with clinical workflows, patient data, and a health ministry context.
What neither of them said: that they had been lucky. That a young analyst had caught it at a dinner table using a training handout from the month before. That this was not a process. It was luck. And luck is not something you can manage.
BrightBridge office. The week after the correction.
The report had gone. The relationship had held. But Amir found he could not stop thinking about the word systematic.
He began applying Putri's checklist to deliverables from the rest of his team. A proposal from a junior consultant: two statistics attributed to industry benchmarks that traced to nothing published. He removed them quietly. Left a note in the margin: Source required before submission — AR. A policy brief from another: a government statistic that was real but three years out of date, presented without a date. He flagged it. Sent it back.
He was catching things. That was the problem. Every catch meant he had been the last line of defence. One person, reading manually, one document at a time. There was no system. There was only him.
He sat with the word and it did not resolve.
A few days later. Evening.
His phone lit up on the desk. Putri.
"Ayah — the programme I attended last month. It's run by P2P Talent Development — under MPC, the Malaysia Project Management Practitioner Community. They do PM professional certification, AI initiatives, ESG management training. They work with local and foreign universities. It's not a vendor course. Part 2 has a new session running next week. Part 1 also has a seat. I think you should go. Both of you."
He read it twice. Then he opened his email and wrote to Dr. Rohini.
"Dr. Rohini, I want to send myself and Maya for the GenAI for Practitioners programme at P2P Talent Development. My daughter attended last month and used what she learned to catch a significant error in a client deliverable before it went out. I'll explain the details in person. Twenty minutes this week?"
"Thursday lunch. David is already coming — our regular quarterly. Faridah has also been invited. It works out — bring the programme details and we'll discuss."
Amir read this twice.
He had asked for twenty minutes. She was giving him a lunch table with the board chairman and the head of operations. And she hadn't told him that was what it was. She had said it works out — the phrasing of someone for whom this was not a coincidence but an arrangement.
He booked the programme before the lunch. Sent Maya the confirmation. Told her only that the dates were set.
Thursday. Ong's Restaurant, rooftop of the BrightBridge office tower, Petaling Jaya.
The lift opened onto a corridor and Amir stepped into the particular quality of air that comes from a kitchen that has been running since morning. He was a few minutes early. He always was.
The restaurant occupied the entire rooftop of the building. Floor-to-ceiling windows ran the full length of the city-facing side, and through them the PJ skyline spread out in the afternoon light — older commercial blocks from the nineties sitting alongside the glass towers of the past decade, a crane visible at the eastern edge still working on something not yet finished. From up here you could see thirty years of PJ's growth at once, layered and unrushed.
Faridah was at the window table, a coffee in front of her. She was fifty-six and had been at BrightBridge for more than twenty years — longer than anyone except Dr. Rohini. She had built the firm's delivery infrastructure from nothing: the quality controls, the submission protocols, the institutional knowledge of which government contact to call when a submission was delayed, which client preferred phone over email. Twenty-five people in Shared Services reported to her. The firm ran, in practical terms, because Faridah knew what she knew.
Beside her, Tan Sri David Lim. Sixty-seven. A retired PwC senior partner — thirty years in the firm, the last fifteen in governance and risk advisory. Four public company boards. The closest professional friend of Dr. Rohini's late father, present at BrightBridge's founding, the institutional memory of what the firm had been built to be. He put his phone away when Amir approached, the deliberate courtesy of a man from a generation that considered the gesture meaningful.
"Amir."
"Tan Sri."
He had asked Dr. Rohini for twenty minutes about a training programme. She had assembled the board chairman and the head of operations at a window table and called it lunch. And she still hadn't arrived.
Something settled in him — not alarm, but the quiet alertness of a man who has learned to trust the feeling that a conversation is larger than it was introduced as. He reached for the menu. He waited.
The lift chimed somewhere behind him.
He heard Dr. Rohini's heels on the floor, and understood, before he turned around, that whatever he had come here expecting was not what this lunch was going to be about.
The Lunch That Changed the Brief
Dr. Rohini reveals an email that has been sitting on her desk for three months. David says one thing, once, and does not repeat it. Faridah asks three questions that nobody answers. Amir asks for a training budget and leaves with a board commission.
GenAI for Practitioner
The verification method Putri used at that dinner table. The governance system Amir was about to discover he needed. Both built live, in the training room, from your actual job.