When employees use AI to assist with their work, who is responsible for mistakes?

At present, many companies encourage employees to use AI tools to assist in daily office work, and AI has gradually become a helper to improve workplace efficiency. However, hidden dangers such as AI illusion, data leakage, and copyright infringement will follow. Once work errors occur, it will easily lead to labor disputes, difficulty in determining liability, and other issues. Experts suggest that companies should refine the list of AI usage scenarios and conduct system publicity and compliance training; relevant departments should issue relevant guidelines to clarify the boundaries of rights and responsibilities.

Recently, the People's Court of Chaoyang District, Beijing concluded a labor dispute case caused by AI-assisted office work. Ms. Wu was terminated from her labor contract because she used external AI tools to process the company's internal documents, so she sued the company to the court.

The reporter's investigation found that many employees are using AI "helpers" to assist in office work, and some employers even actively encourage it. However, improper use of AI tools is also prone to various risks. Once a work error occurs, employees may be punished for their mistakes or even fired.

Some companies encourage employees to use AI to work

"I often use large domestic models when writing materials to help me collect data, formulate article outlines, and optimize sentence expressions." Li Fei (pseudonym), who works in a middle and back-office position at a bank in Beijing, told reporters that he uses AI tools mainly for ordinary, non-confidential word processing.

"The unit encourages us to use AI, and will regularly form and publish "AI + Financial Work" briefings to demonstrate the effectiveness of the application." Li Fei said that the bank also provides a lot of convenience for employees to use AI.

"We have a dedicated AI working group, and have also provided employees with several mainstream large models and preset multiple Agents, involving text processing, table analysis, PPT production and other demand scenarios, and will also organize relevant training and competitions," Li Fei said.

The reporter's interview found that, similar to Li Fei's bank, many enterprises and institutions have begun to encourage employees to use AI to assist in office work, and provide various ways to support employees in learning and exploring new AI tools.

"In order to encourage everyone to use AI to improve programming efficiency, the company can provide employees with a certain amount of token usage every month, and the excess amount will be reimbursed." Mr. Wen, a programmer working at an Internet company in Hubei, told reporters, "Even if the model iterates quickly, the cost is relatively high, and many Internet companies are willing to let employees try as many as possible."

"We clearly encourage the use of AI at the department level, and older employees will also be required to learn." Ms. Zhang from Beijing is a practitioner in the publishing industry. She and her colleagues often use AI tools to design the layout of posters and product content.

"Currently, many companies encourage employees to use AI tools, which have three main benefits." He Ping, a professor at the Law School of Wuhan University of Technology, told reporters. The first is to improve efficiency and free employees from repetitive tasks; the second is to improve capabilities. Ordinary employees of small and medium-sized enterprises can also get close to professional level assistance by using AI tools; the other is to optimize management. Some companies deploy AI models locally, which is equivalent to incorporating AI into controllable organizational processes.

Risks of abusing AI tools in the workplace

"Do you need me to fine-tune some of the sentences?" In October this year, a reader found text that seemed to be an AI prompt in a published paper novel, and took a picture and posted it on social platforms, which attracted widespread attention.

In the same month, a set of popular science books contained an erroneous picture of a "two-headed shrimp" in the drawing and was removed from the publisher's website. This set of books states in the preface that "most of the pictures in this set of books are generated by AI drawing tools."

In the technology industry, there are also situations where AI tools can go wrong. "If you directly use AI to write code, once the instructions are not clear enough, you may make up non-existent "functions" or even modify the files in the computer." Mr. Lin, who works in a technical position at a chip company, told reporters.

"There are many potential risks in using AI assistance at work." Zhou Yuan, an assistant professor at the Beijing Institute of Technology Law School, told reporters that if employees input non-desensitized information containing customer privacy and financial data into a third-party public large model, it may lead to the leakage of trade secrets and data; if the pictures, codes or copywriting generated by AI are used to induce the reproduction of the original work, it may also infringe the intellectual property rights of the third party.

"In addition, AI large models may have "illusions" and generate seemingly reasonable but erroneous information. If employees fail to fulfill their review obligations and directly use it for external release or delivery to customers, it may lead to errors in work results and cause losses to the employer. They may be liable for breach of contract, damages, or even be unilaterally dismissed for fault," Zhou Yuan said.

So, how should employees standardize the use of AI to assist in office work?

"AI-assisted office work cannot be used casually. You should first confirm whether the company has regulations on AI use, data security, confidentiality management, etc." He Ping reminded that when it comes to internal data processing, employees should try to obtain retainable authorizations, such as written confirmations from emails, work groups, and office systems, etc.

"During use, core secrets, undisclosed data and confidential documents should not be uploaded, and personal information of customers and employees should not be entered at will. After use, pay attention to saving system notices, training requirements, work instructions and authorization records, which can be used as evidence in the event of a dispute." He Ping said.

Who is responsible for mistakes?

In an AI-assisted office labor dispute case heard by the Chaoyang Court, the client, Ms. Wu, worked as a legal counsel for an insurance company. On the recommendation of the company leader, she used AI tools to process internal documents. The company considered her to be a "serious violation of the company's information security and other relevant regulations," and she was terminated from her labor contract. After hearing the case, the court ruled that both parties should continue to perform the labor contract, and the company should pay the difference in year-end bonuses, performance wages, wage losses, etc.

"In this case, the court clarified the reasonable boundaries for employees to use AI to improve quality and efficiency." Zhou Yuan analyzed the case and found that Ms. Wu was recommended by her leader to use AI tools for the subjective purpose of optimizing work; objectively, the company did not have clear prohibitive regulations on AI-assisted office work, nor did it conduct compliance training; as a result, no actual leakage losses were caused, so it did not constitute a "serious violation of rules and regulations."

How to avoid problems with AI-assisted work? Once a problem occurs, who should bear the responsibility?

"The unit has issued a system document that regulates the use of large models, with a clear list of usage scenarios and confidentiality requirements. When we use AI to work, we basically know what can and cannot be done." Li Fei said that the unit will also have relevant departments to carry out inspections to ensure that implementation is in place.

"Enterprises cannot just require employees to use AI to improve efficiency, but turn the lack of their own management systems into unilateral risks for employees." Zhou Yuan reminded that enterprises should improve rules and regulations, update employee handbooks or issue AI-assisted office compliance guidelines to clearly list "scenarios where AI can be used" and "scenarios where AI is disabled."

"In terms of information security, it is not enough to generally write "disclosure of confidential information is prohibited." AI large models are emerging tools. As managers, companies need to establish clear data classification directories and operational red lines specifically for AI scenarios." Zhou Yuan said.

"Enterprises should also pay attention to the disclosure and training of rules and regulations. It is necessary to clarify the principle that "enterprises must not make it clear in advance and shall not be held responsible afterwards." That is, if an enterprise fails to formulate and disclose AI usage specifications through democratic procedures or conduct training, it shall not be deemed to have serious disciplinary violations solely on the basis of employees uploading information." He Ping said that this not only forces enterprises to improve their systems, but also prevents unreasonable dismissals on this basis.

"At present, my country's labor law system has a certain lag in regulating the management of artificial intelligence algorithms. Special guidance or typical case guidelines should be issued as soon as possible for "employees using AI-assisted office work." Zhou Yuan suggested that in combination with the relevant provisions of the Labor Contract Law and the Civil Code, the definition standards of "serious dereliction of duty causing significant damage to the employer" for AI-assisted office work should be clarified in judicial practice, and the systemic risks of the tools provided by the enterprise and the personal faults of employees who have not been strictly reviewed should be fairly divided, and the safety of technology and the efficiency of use should be balanced.

Reporter Zhang Xiaojian