From Panic to Profit: A Blunt Guide to Investigations in the Age of AI
Worried about the next harassment complaint in your inbox? You should be. One of the fastest ways to increase legal costs is to mishandle an investigation, allow relevant messages to disappear, or lack a documented process. Once litigation is reasonably foreseeable, a court may impose consequences for failing to preserve relevant evidence.
The practical approach is straightforward: preserve the evidence, conduct a fair and credible investigation, and use modern tools without allowing them to make the judgment calls. AI can help sort information and prepare summaries, but people must remain responsible for evaluating credibility, resolving factual disputes, and reaching legal conclusions. That is how sound legal risk management can support informed business decisions and protect the bottom line.
What Changed, and Why You Should Care
Federal AI policy has shifted considerably. Federal agencies removed or revised several AI-related employment resources in 2025, leaving employers with fewer ready-made instructions and more responsibility for developing sound internal processes. Title VII and other applicable employment laws still apply when employers use AI-assisted tools in hiring, performance management, discipline, termination, or other employment decisions.
The message for business leaders and HR professionals is simple: an employer cannot point to a software tool and disclaim responsibility for the result. The employer remains accountable for its decisions and for the process used to reach them.
In January 2026, the EEOC rescinded its 2024 Enforcement Guidance on Harassment in the Workplace. That action did not eliminate federal laws prohibiting unlawful workplace harassment or the Supreme Court decisions addressing employer liability. Employers still need effective complaint procedures and must respond promptly and appropriately when harassment concerns arise.
Those measures remain central to the employer defense recognized by the Supreme Court in 1998. What has changed is the level of detail in federal guidance. As a result, an employer’s written policies, training, response procedures, and investigation practices carry even more weight.
The Move That Can Save You Money
Preserve relevant chats promptly. When a complaint makes litigation reasonably foreseeable, assess whether a legal hold is appropriate, identify the people and systems that may contain relevant information, and suspend automatic deletion settings when necessary.
Microsoft Teams content may be discoverable. Microsoft provides tools for preserving, collecting, reviewing, and exporting Teams content through Purview eDiscovery. Without appropriate preservation measures, an employer may lose earlier versions of edited messages, reaction histories, meeting recordings, and other information associated with a conversation.
If your team cannot locate a Teams conversation, opposing counsel may still find it. An employee or other participant may have retained a copy, taken a screenshot, forwarded the message, or preserved it on another device.
Understand the spoliation rule. When electronically stored information that should have been preserved in anticipation of or during litigation is lost because reasonable preservation steps were not taken, Federal Rule of Civil Procedure 37(e) may permit remedies if the information cannot be restored or replaced through additional discovery and the loss prejudices the other side. When a party acted with an intent to deprive another party of the information’s use, the rule permits more serious sanctions.
The preservation analysis may need to include direct messages, channel posts, meeting recordings, reactions, attachments, shared drives, text messages, and other side channels, depending on the facts. State courts, including courts in Texas, apply their own rules and standards concerning preservation and spoliation.
Use AI Like an Associate, Not an Oracle
AI can assist with discrete investigation tasks, such as preparing draft interview summaries, creating evidence tables, identifying recurring topics, and reorganizing notes. It works best when the assignment is specific and the output is reviewed by a person who understands the matter.
One practical prompting method is RISEN, which stands for Role, Instructions, Steps, End Goal, and Narrowing. The structure helps produce focused, usable results. It is not a substitute for legal judgment.
One rule comes first: use an approved enterprise tool controlled by the company, not a free public chatbot. Investigation files often contain sensitive personal information, confidential business material, and legally protected communications. Before using an AI tool, confirm the organization’s data-use terms, retention settings, access controls, and contractual protections.
Consider a common failure: an AI-generated summary reverses a witness’s timeline and states that an event occurred three months after the employee’s hire date when it actually occurred three months before the interview. That is not a harmless drafting error. A date error can undermine a credibility assessment and compromise the investigation’s conclusions.
The remedy is simple, even if it is not exciting: verify before inserting. Check every name, date, quote, and material assertion against the source record. Treat prompts, AI outputs, and draft summaries as investigation materials that may be subject to discovery, depending on the circumstances.
A practical assignment might ask an AI tool to organize evidence into four categories: complainant, respondent, witnesses, and documents. The tool could then identify areas of agreement and conflict. That structure may be useful, but every entry must be checked against the underlying record before it appears in an investigative report.
Harassment Liability in One Paragraph
The Supreme Court’s decisions in Faragher v. City of Boca Raton and Burlington Industries, Inc. v. Ellerth established important principles governing employer liability for harassment by supervisors.
For this purpose, a supervisor generally is someone the employer empowers to take tangible employment actions, such as hiring, firing, failing to promote, or making a significant change to an employee’s duties or benefits. When supervisor harassment results in a tangible employment action, the employer may face vicarious liability. In certain hostile-work-environment cases that do not involve a tangible employment action, an employer may have an affirmative defense based on reasonable efforts to prevent and promptly correct harassment, together with the employee’s unreasonable failure to use available preventive or corrective opportunities.
The practical point remains the same: maintain effective complaint procedures, respond promptly, investigate appropriately, and take corrective action when warranted. State law may impose additional obligations. In Texas, for example, employers may face liability for sexual harassment when the employer or a supervisor knew or should have known about the conduct and failed to take immediate and appropriate corrective action. Depending on the circumstances, supervisors may also face individual liability.
Technical Takeaways You Can Apply This Week
Build a repeatable prompt kit. RISEN provides a practical structure: assign a role, give clear instructions, list the steps, state the end goal, and narrow the requested tone, length, and audience. Break the work into discrete assignments. Focused prompts generally produce more usable results than vague, all-purpose requests. Save approved prompts so the team can use them consistently.
Make “summarize, verify, insert” the rule. Use that sequence for chat threads, meetings, and interview notes. Verification means checking every name, date, and quote against the source record before including it in a report or decision memo.
Expect fabricated details and design around them. AI tools can produce incorrect information with unwarranted confidence. Build verification steps into the workflow, and do not allow AI to make credibility determinations or reach legal conclusions on its own.
Reconstruct the conversation. Context matters. A single message removed from a longer exchange may appear more serious—or less serious—than it was. Use available eDiscovery tools to review conversations in context rather than as disconnected lines of exported text.
Keep AI governance practical and documented. The NIST AI Risk Management Framework and its generative AI profile offer voluntary concepts for identifying and managing AI-related risks. Because NIST continues to update its materials, confirm that your organization is using the current version. Those concepts can help HR and Legal document data sources, assess risks, limit access, and maintain meaningful human oversight.
Remember that AI does not displace civil rights law. Changes in agency guidance do not eliminate federal or state prohibitions on discrimination. Hiring tools, employee-monitoring systems, and workplace analytics still must be used consistently with applicable employment laws. Some states impose additional requirements. Illinois regulates the use of AI in employment decisions, and Texas law prohibits using AI with an intent to unlawfully discriminate. Those legal limits should remain part of any AI governance process.
Monday Morning Checklist for HR and Operations
- Determine whether a preservation obligation exists. When appropriate, issue a written legal hold covering relevant Teams chats, meeting recordings, channel posts, direct messages, attachments, and shared drives. Coordinate preservation with IT. Include personal phones and text messages when employees use them for work, and document when the hold was issued and to whom.
- Use an intake form that identifies everyone who may have seen or heard the conduct and every system that may contain related information. Record when the company first learned of the concern.
- Use the RISEN method to prepare interview questions, but check all quotations and dates against interview notes or transcripts before finalizing summaries.
- Organize the draft report into four categories: complainant, respondent, witnesses, and documents. Conclude with areas of agreement and conflict. Counsel should review the legal analysis against the applicable framework.
- Remind everyone involved in writing that retaliation is prohibited.
- If the complaint involves national origin, review the EEOC’s national origin discrimination resources and compare the organization’s policy and process with applicable legal requirements. National origin protections apply to all workers, including American workers.
The Business Case, Without the Fluff
Every hour responsibly removed from an investigation is an hour managers can return to running the business. Careful evidence handling can reduce discovery disputes and spoliation issues later. A clear report grounded in the record gives counsel a more reliable foundation if the matter becomes an agency charge or lawsuit.
That is the business value of disciplined investigation practices: fewer avoidable errors, more reliable decision-making, and a process that can withstand scrutiny when the stakes rise.
What We Do for Employer Clients
Treaty Oak Employers’ Law Group is a Texas law firm with attorneys licensed in Texas, North Carolina, Wyoming, and Colorado. The firm represents employers, develops practical policies, conducts workplace investigations, assists with preservation and eDiscovery workflows, and trains HR professionals and managers on responsible AI use without transferring judgment to a software tool. When disputes escalate, the firm represents employers in the resulting proceedings.
If AI is already entering your investigation process, the process should account for how the technology is being used. A sound workflow can take advantage of AI-assisted tools while keeping legal judgment, factual evaluation, and accountability with people.
