AI Changed the Job. The Paperwork Didn’t Notice.
Artificial intelligence changed how many employees perform their jobs over the past year. Job descriptions, exemption analyses, confidentiality policies, and personnel files often did not keep pace. That gap may remain unnoticed while operations run smoothly. It becomes significant when an accommodation request, termination, wage claim, or discovery request requires the company’s records to explain how work is actually performed.
Other Duties as Reassigned
At many companies, an employee began using an AI tool to complete part of a job. The employee’s output increased, turnaround time decreased, and the arrangement appeared to work. Because the tool was inexpensive, required no installation, and could be adopted immediately, there may have been no procurement process, approval, or written policy.
That is not necessarily a management failure. It is a predictable result of making new technology easy to adopt. The same initiative that helps a company grow can also lead employees to try a tool before anyone has considered its effect on job duties, wage classifications, confidentiality, or recordkeeping.
The tool itself may not be the central problem. The problem is that the organization chart changed quietly while the documents describing the organization remained the same. That is a paperwork issue today. It can become a legal issue the first time someone outside the company reviews those documents.
Three Things That Break When a Task Moves
1. The job description no longer matches the job.
Job descriptions identify essential functions, and essential functions inform the reasonable-accommodation analysis under the Americans with Disabilities Act (ADA). If the written description still lists tasks that software now performs, but does not identify the employee’s review, verification, or judgment responsibilities, the company may be relying on a document that describes a position no one currently holds.
That mismatch matters when an employee requests an accommodation. If the company denies the request because the employee cannot perform a task that the software now handles, the company may have difficulty explaining why that task remains an essential function. The issue often surfaces after the accommodation has already been denied, when the company is asked to justify its decision using records that do not reflect current practice.
2. Restructuring can appear to be a cover story.
A restructuring decision can begin to look like pretext when the business reason was not documented at the time. Administrative and support positions may be affected by automation, and those positions may have demographic characteristics that create additional discrimination risks. Eliminating or consolidating a role may be lawful. Making the decision without a contemporaneous explanation, then assembling the rationale after a charge is filed, can make an ordinary operational decision appear suspect.
“We no longer needed the role” is a conclusion, not a business record. The company’s file should explain what changed, why the work could be reassigned or eliminated, who made the decision, and when the decision was made. That documentation should exist before a claim arises.
3. Overtime exemptions can quietly fail.
An exempt classification can become inaccurate when an employee’s duties change. The duties test examines the work an employee actually performs, not simply the employee’s title. If an exempt employee’s substantive work moves to an AI tool and the employee is left primarily with oversight, data cleanup, or administrative tasks, the exemption the company has relied on for years may no longer apply.
Misclassification can result in unpaid overtime, liquidated damages, and a lookback period that may extend to three years for willful violations. When several employees perform substantially similar work, the potential exposure may extend beyond the first person to raise the issue.
The Wage-and-Hour Problem Nobody Puts on the Agenda
Productivity expectations often adjust to the fastest adopter on a team. Once one employee produces work at three times the prior pace, that output can become the informal standard for everyone, including nonexempt employees who were never told which tools they could use or how their time should be recorded.
A nonexempt employee trying to meet that standard may use a personal account, a personal phone, or an AI tool after hours. The employee may not record the time. If the employer knows, or has reason to know, that the employee is performing work, that time may be compensable.
The activity may remain invisible until a former employee’s lawyer requests application history, device records, or other evidence of work performed outside normal hours. At that point, an informal productivity practice can become a documented pattern.
The Confidentiality Exposure in the Chat Window
Employees who find an AI tool useful may use the free or personal account already available to them. They may do so without reviewing the account’s terms of service, which could permit submitted information to be retained or used to improve the underlying system.
For a company that holds client files, patient information, salary data, or investigation materials, that creates confidentiality and compliance concerns. For a company that claims certain information is a trade secret, the consequences may be greater.
Protection under the Texas Uniform Trade Secrets Act and the federal Defend Trade Secrets Act depends in part on whether the company took reasonable measures to maintain secrecy. If the company cannot show that it took reasonable steps to protect information, a court may question whether the information qualifies as a trade secret.
Pasting a pricing model into a public chat window may undermine that position. If no policy prohibited the conduct because the company had not adopted a policy, the absence of a rule may not appear neutral. It may instead suggest that the company never addressed how confidential information should be handled.
What Texas Requires Now
The Texas Responsible Artificial Intelligence Governance Act (TRAIGA), House Bill 149, took effect on January 1, 2026. The law is narrower than some earlier proposals. It prohibits developing or deploying an AI system with the intent to unlawfully discriminate against a protected class. Disparate impact alone is not sufficient to establish discriminatory intent under the statute, and earlier impact-assessment requirements were removed.
Enforcement primarily runs through the Texas Attorney General. Employees do not have a private right of action under TRAIGA, and the Attorney General must provide notice and a 60-day opportunity to cure before seeking penalties.
For many companies, the important point is not only the prohibition itself but also the law’s approach to compliance and documentation. TRAIGA recognizes substantial compliance with the NIST Generative AI Profile or another nationally or internationally recognized AI risk-management framework in connection with its liability provisions. If the Attorney General investigates, the company may need to describe the system and produce relevant documentation identified by the Attorney General.
A company with written AI governance materials has records to produce. A company without them may have little more than an informal message thread and employees’ differing recollections.
The analysis becomes more complicated when a company has employees in other states. For example, Colorado Senate Bill 26-189 is scheduled to take effect on January 1, 2027, and its requirements differ from the Texas approach. Colorado’s law covers certain AI tools that materially influence hiring and other employment decisions involving Colorado applicants and employees. It also includes notice, post-decision disclosure, and record-retention requirements.
A single policy built to address the most demanding applicable requirements may be manageable. A Texas policy quietly applied to Colorado employees may become an exhibit in later litigation.
What Unverified Output Costs the People Standing Next to It
Gauthier v. Goodyear Tire & Rubber Co., a wrongful-termination case in the Eastern District of Texas, illustrates the cost of unverified AI output. The former employee’s lawyer filed a brief citing two nonexistent cases and including quotations that did not appear in the cited sources. The material had been generated by an AI tool and was not adequately checked. The court sanctioned the lawyer $2,000.
Goodyear reported spending $7,521.26 on a reply brief that required its lawyers to investigate the nonexistent cases and address the resulting issues. Even after Goodyear identified the false citations, the lawyer did not fully review the work until the court ordered an explanation.
The company that did not misuse AI still had to spend money addressing the consequences. That pattern applies more broadly. Unverified output is often most costly for the person or organization responsible for correcting it, not for the person who first generated it. Inside a company, that usually means the employer.
The court’s response provides another useful lesson. It did not prohibit AI use. Instead, a judge in the same district later required filings to state whether AI was used and to certify that someone reviewed the output.
The same principle applies when an employee uses AI to prepare a client proposal, report, analysis, or other work product. Verification should be a documented responsibility, not an assumption.
Five Steps Worth Taking Before Year-End
- Adopt a written policy. Identify approved tools, prohibited categories of information, verification requirements, and disclosure obligations. For example, employees may be required to tell their managers when AI contributed to a work product. A short, dated policy is more useful than no policy at all because it shows that the company addressed the issue.
- Update job descriptions. Revise the descriptions for roles whose work has changed. Include review, verification, judgment, and oversight responsibilities where those duties are part of the actual job.
- Document business reasons. For any role that is reduced, combined, or eliminated, document the business reason before the decision is implemented. Include the operational changes that led to the decision rather than relying on a general statement that the role was no longer needed.
- Recheck exempt classifications. Review classifications when duties move from an employee to an AI tool or when a position becomes primarily focused on oversight, cleanup, or administrative work.
- Use a risk-management framework. Adopt and periodically review a recognized AI risk-management framework, and keep records showing when and how it was adopted. Under TRAIGA, substantial compliance with a recognized framework may be relevant when a company identifies and addresses a potential violation through its internal review process.
These steps do not require a company to stop adopting useful technology. They also do not require the company to characterize past decisions as mistakes. Employees adopted tools because the tools were inexpensive, accessible, and effective. The documents now need to catch up with the way the work is actually being performed.
AI does not need permission to change how work gets done. Once the work changes, however, the documents describing it should change as well. Treaty Oak’s HR process audit reviews potential gaps involving job descriptions, classifications, hiring and separation practices, and other HR processes before an outdated document is asked to explain a job that no longer exists.
