(By You Yunting) Imagine this scenario: you have worked at a company for five years, and your daily emails, meeting minutes, and handled contracts are all stored on the company’s servers. One day, you are called in for a talk and HR informs you that your position is redundant. Your replacement is a “Skill file” distilled from your five years of work records—an AI system that can perform your job duties using this file. Today, we will discuss whether such skill distillation raises legal issues.
With the rapid development of AI, Anthropic introduced the Agent Skills open standard (the SKILL.md file) in late 2025, providing a standardized container for packaging human expertise. Using this standard, companies can employ AI technologies to analyze employees’ work documents, emails, chat records, and even meeting recordings and extract and compile into machine-readable Skill files their experience and workflows formed at work. By accessing these Skill files, AI systems can perform tasks that traditionally relied on human experience, acquiring skills such as contract review, tax filing, and client communication.
I. Currently Lawful Distillation Practices
From a legal standpoint, employers do have a basis to claim rights over employees’ work-related outputs. Copyrighted works created within the scope of their duties are considered service works: for ordinary service works, employees may authorize others to use them two years after their resignation; for special service works, their rights belong directly to the employer. Service inventions, under patent law, also belong to the employer. As for general work skills that do not fall into these categories, the employer’s summarization and extraction of such skills currently face relatively limited legal controversy. In short, as long as the materials used are lawful work documents and outputs, skill distillation itself possesses a certain degree of legitimacy.
II. The Red Line: Privacy and Personal Data Protection
The core legal controversy centers on privacy and personal data protection:
First is the source of the data. The device may belong to the company, but the information may be not necessarily. Personal chat records or private emails stored on work devices—even if physically located on company servers—still fall within the scope of personal privacy. Training AI with such data may infringe upon the employee’s personal rights. A more subtle issue is that many employees consent to the monitoring of their company devices when they are onboarded. However, this is fundamentally different from consenting to the use of their behavioral data to train an AI system that may replace them. The former cannot be interpreted as the latter.
Second is the content of the generated output. Distilled “Skill files” must not contain employees’ biometric identifiers, such as voiceprints, facial features, or habitual expressions. Such data is classified under most jurisdictions as sensitive personal information, requiring separate and explicit consent. Using an employee’s likeness or voice for a “Digital Human” or any commercial services without authorization may violate his/her portrait and voice rights.
Data security is also a significant concern. Under the current regulatory system, there are almost no constraints on where Skill files are stored, who has access, or whether they are transferred across borders or shared with third parties. Employees have little ability to track the flow of their digital twin.
III. Power Imbalance in Employment Relationships
From the perspective of employment relationships, the process of distilling professional skills into Skill files involves structural unfairness. On one hand, skills accumulated over years are distilled into replicable digital assets within a short time. Employers lose their market pricing power but cannot share the resulting benefits. On the other hand, the current copyright law protects expressions rather than ideas or skills. Employers can hardly claim rights over extracted skills, resulting in a “legal protection vacuum”.
More concerning is that some companies may quietly record employees’ work to extract Skill files, and then use them as a basis for layoffs—employees have inadvertently provided the material which determines their own job retention. Furthermore, even if employees become aware of the infringement, proving it is extremely difficult. Distilled skills are not preserved as complete files. With only clues in prompts, there is a very high threshold to provide preliminary evidence in court.
IV. Legislation and Judicature Regarding AILagging Behind Due to International Competition
Currently, no unified and stable rule system has yet been established worldwide regarding whether AI training constitutes copyright infringement or requires prior licensing. The slow pace of legislation and judicature has left a significant gray area for distillation.
A deeper dilemma lies in the fact that this technological wave can hardly be blocked by any single country’s legislation. If one country takes the lead to adopt strict compliance requirements, it may slow down its industrial development in the short term, allowing competitors in other countries to gain advantages. For example, while parts of the U.S. gaming and film industries are currently resisting AI-generated content, I am not optimistic about the prospects of such resistance. In fact, the skills of these professionals have already been distilled through large-scale training content—including images, videos, music, and performances, which AI can already create in place of human to a certain extent. Therefore, it is foreseeable that if U.S. industries refrain from using AI, practitioners in other countries will gain a competitive edge, ultimately forcing U.S. industries to adopt AI as well.
For this reason, I am very concerned about the job prospects of young people. In the mobile internet era, efficiency improvements driven by technological progress have already pushed many graduates into jobs such as delivery services and ride-hailing services. In the AI era, this trend may intensify, as AI-driven efficiency improvements may eliminate more traditional white-collar jobs. In particular, the emergence of Skill files may erode entry-level positions, weakening the base of the current labor market pyramid and potentially leading to a talent gap in the future.
As this future is already foreseeable, more reasonable and forward-looking institutional designs in legislation and judicature should be advanced. The direction may not be complex: first, to establish a system of informed consent for skill distillation, prohibiting the use of standard clauses as a substitute for genuine authorization; second, to refer to the incentive logic for service inventions, allowing employees who has provided training material for Skill files to share the benefits generated by AI assets. Technological progress cannot be stopped, but the distribution of benefits can be shaped by legal institutions.
Lawyer Contacts
Disclaimer of Bridge IP Law Commentary
Short Link: