The effectiveness is not yet seen, the boss has a headache because the employee "reported harm" due to AI

(Dan Tri) - The code crashes the entire system, from unfeasible plans to incorrect quotes... AI can help employees work faster but also makes superiors tired of having to bear the consequences.

One line of code from AI, the whole project takes a week to fix

The AI-generated code, not yet fully tested, was posted directly to the official operating system by an employee.

Not long after, the system stopped working and a large amount of the project's data "evaporated".

The remaining members of the group had to gather together to find the cause and fix the errors that occurred.

But after nearly a week, the problem still could not be completely fixed, customers continued to complain.

As a result, the whole team was forced to return to the original version, meaning that much of the time and effort spent fixing previous errors was almost in vain.

That is the recent story at a company specializing in technology solutions headed by Ms. Nguyen Thi Trang.

Ms. Trang believes that AI can help programmers find solutions faster, create source code drafts, check for errors or shorten repetitive operations.

“The problem appears when users consider the AI ​​results as a finished product, instead of an option that needs to be evaluated,” she said.

In the programming department, in addition to the above case, Ms. Trang also noted that many employees let AI create code and then update it directly to the system without checking carefully.

Not only that, in the business analysis department, risks also continue to appear.

"The anti-censorship staff submitted content created by AI, which on the surface was quite complete in terms of text, presented coherently but easily repeated ideas, overlapped parts and lacked logic throughout," the female director assessed.

Ms. Trang affirmed that AI itself is not the cause. The bigger problem lies in employees using the tools incorrectly.

Some users do not know how to make appropriate requests, causing AI results to not be close to the business context. Others copy almost entire content or source code without checking.

The reaction of employees at the company when consequences occur also makes handling more complicated.

"A few even tend to silently correct errors, avoid communication or only admit when the manager reviews the entire process," Ms. Trang worries.

A series of errors due to employees using AI "novices"

Besides Ms. Trang's case, Mr. Tien Duc, a tourism service business operations manager, is also frustrated by the way his employees use AI.

According to Mr. Duc, a part of workers is forming the habit of relying on "asking AI for everything", including tasks that must be based on practical experience.

He once witnessed an employee using AI to build a schedule for a group of guests and then using almost the same content created.

Looking at the plan, the program is presented quite reasonably, with full travel times, attractions and accompanying services.

But upon closer inspection, a series of problems appeared: the travel time was too short, some attractions closed on the day the group arrived and some services had not been confirmed with partners.

Unfortunately, the schedule has already been sent to guests. The business must then readjust the entire program, contacting customers, bus operators, hotels and suppliers.

Some items have been pre-ordered, so additional change costs may arise. Staff also had to spend a lot of time explaining to customers to accept the new plan.

For Mr. Duc, the biggest loss is not necessarily in the amount of money incurred but in the trust of customers.

More worryingly, in order to quickly create reports, he also accidentally witnessed an employee directly giving customer lists, emails, phone numbers and a series of sensitive data to AI tools.

The situation of employees using AI irresponsibly is also happening in the quotation department of a business providing products and services.

Ms. Thu Ha, the person in charge of the department, said that employees often use AI to compose price offers, product descriptions, create plan comparison tables or write emails to customers.

But Ms. Ha also saw many cases where the tool was asked to suggest prices and discount rates, even though the AI ​​did not know input prices, shipping costs, minimum profits, sales policies and individual agreements with each customer.

"There was a time when an employee sent the customer a price quote directly in which some products had the wrong unit price," Ms. Ha said.

And then only when the customer confirmed the order did the accounting department discover that the selling price was lower than the allowed level.

AI runs first, management follows

In fact, besides the above cases, incidents caused by employees using AI without control are also common globally.

Typically, in March 2026, Meta confirmed that an engineer had implemented a suggested solution given by the AI ​​agent to handle a technical issue on the internal forum.

This action exposed a large amount of sensitive user and business data to engineers for approximately two hours.

Another case is that during an investigation by the Chairman of the US Senate Judiciary Committee in 2025, two federal judges admitted that personnel in their office used AI to research and draft court documents.

Because the review process was not fully implemented, documents full of AI errors were issued and had to be withdrawn immediately afterwards.

In addition to errors caused by employees and censorship processes, another problem is the situation of employees blatantly providing internal data to AI.

For example, Harmonic Security's research on one million commands and 20,000 files sent to 300 AI engines in the second quarter of 2025 found that up to 4.4% of commands and 22% of uploaded files contained sensitive information.

Source code is the most common type of data posted by workers to chatbots, along with sales plans, financial models, investor information, legal strategies, internal emails and customer data.

In particular, a large proportion of data is sent via personal accounts or free AI versions such as ChatGPT, Gemini.

As a result, businesses face a series of risks such as leaking business secrets, legal liability and reputational deterioration.

Faced with the above situation, many experts recommend that companies need to soon develop clear regulations for using AI and tighten the approval process for jobs.

In parallel with that, deploy an internal AI platform when there are enough resources or require employees to only use approved tools and have transparent information policies.