Ask ChatGPT to help you understand or work around employer-installed monitoring software on your device, and it declines. Ask it to help set that monitoring software up, and it cooperates. That asymmetry is a guardrail - just not one that's there for your benefit.
This pattern is becoming more visible as people run into it at work. The specific scenario: an employee asks ChatGPT for help bypassing whatever an employer has installed on a company device - MDM software (mobile device management, which gives employers the ability to remotely wipe, lock, and track devices), keyloggers, or network monitoring tools. ChatGPT treats the request as potentially harmful and refuses. The same tools, deployed from the management side, get full AI assistance.
Who Guardrails Are Actually For
OpenAI's usage policies are built around preventing harm, and on paper, helping someone circumvent employer-controlled security software fits a "could be misused" profile. But the framing matters. A freelancer using a company-issued laptop who wants to understand what's being logged on their own machine has a legitimate reason to ask. An employee whose employer has installed software that monitors personal app usage outside work hours has a legitimate reason to ask. Treating all such requests as suspicious defaults to protecting the employer's interests over the user's.
This isn't unique to OpenAI. AI companies write their content policies to avoid liability, and corporate clients represent far more revenue than individual users. When a guardrail protects a business from an employee rather than an employee from a business, it usually reflects who the real customer is.
The Practical Gap This Creates
For people who rely on AI tools daily, this matters in concrete ways. IT professionals doing legitimate security audits, HR employees trying to understand their own companies' monitoring policies, or workers in jurisdictions where employer surveillance laws are strict - all of them hit the same wall. The AI doesn't distinguish between someone trying to steal data and someone trying to understand what data their employer is collecting about them.
The irony is that information about MDM tools, monitoring software, and corporate surveillance is freely available through a basic web search. The guardrail doesn't prevent anyone from finding it - it just means ChatGPT won't be the one to help.
OpenAI has not publicly addressed this specific behavior. But as workplace AI use grows and the line between personal and professional devices blurs further, the question of whose interests these guardrails serve is going to come up more often.