Rod

HR asked us to build an internal AI chatbot that could answer employee questions about company policies. We connected it to the employee handbook, benefits documents, and twelve years of internal company announcements. Named it AskHR. It worked beautifully. Employees could ask about PTO, health insurance, parental leave, whatever. Then someone asked: "Why haven't we received raises in two years?" The chatbot responded: "According to the CFO's March 2025 announcement, the company is experiencing financial constraints." Then it helpfully added: "However, the company approved $340,000 in executive bonuses during the same period." HR called me immediately. "WHY IS IT SAYING THAT?" I explained that both announcements were in its knowledge base. They asked me to remove the bonus information. I asked if it was inaccurate. "No, but it's inappropriate." Then someone asked AskHR why the company stopped providing free lunches. It explained the decision was made to reduce operating expenses. Then cited a leadership announcement celebrating our new $180,000 executive retreat. Employees started asking questions we'd never anticipated. "Why does the CEO have a company car?" "Who approved the private jet expense?" "Why was the employee wellness budget cut?" AskHR answered every single one. With sources. By Thursday, HR had received 200 complaints. Not about the chatbot. About the company. The CEO demanded we shut it down. Said the system was spreading negativity. I explained that it was literally repeating information leadership had published themselves. He said we needed to make it more aligned with company values. I asked which values. He said he'd get back to me. We spent $140,000 building an AI that could accurately explain company policies. It lasted four days. Turns out the one thing leadership didn't want employees having access to was leadership's own decisions.