What can you safely tell AI at work?
When Bhaskar Bharat Sawant troubleshoots a software problem, an error message often provides the first clue. It can also present a dilemma: how much of that message can he safely share with AI tools?
Sawant, a lead engineer and solution architect at building-products manufacturer Cornerstone Building Brands, works with legacy applications and newer technologies. GenAI can help explain an error or suggest a way forward. But the input it needs to understand the problem may expose personally identifiable information, a customer’s competitive secrets, or his own company’s proprietary information.
“Logs or error messages can reveal internal service names, configuration details, or business processes,” Sawant said. “You have to protect that information whether it belongs to your employer or a customer.”
A customer’s name is an obvious identifier. A method name, file path, or sequence of requests between systems is less conspicuous. Together, those details can expose how an application works or how a business operates — even when no individual can be identified.
Sawant’s approach is to construct a smaller, fictional version of the problem. “I create a small code example with made-up data that reproduces the issue,” he said. “Then I ask myself whether someone outside the company could piece together proprietary information from it. If they could, I simplify it further or keep it out of the AI tool.”
That process takes judgment and effort before AI can save him any time. It also exposes the difficulty behind a simple workplace instruction: don’t share confidential or sensitive information with generative AI. Employees must translate that warning into decisions about actual work, often while trying to meet a deadline.
What remains after you remove the names
Employees can recognize the need to protect information and still overlook what they are sharing. Julian Gage, who provides fractional data protection officer services through Engage Compliance, recalls advising a health technology company where an employee uploaded a spreadsheet containing treatment notes and clinic customer numbers to an AI service that was not company-approved. The records did not name the patients, but each customer number provided a link back to an identifiable person.
“Replacing a name with a customer number doesn’t make the information anonymous,” he said. “Someone with access to the clinic’s records could still establish whose treatment notes they were.”
Meeting transcripts present a similar problem. Removing a colleague’s name may accomplish little if the discussion describes an absence, medical treatment, or other circumstances that make the person recognizable.
“You have to look at the content, not just the names,” Gage said. “If people can connect the dots and identify the individual, you haven’t made the information anonymous.”
Attachments introduce further opportunities to miss something. Screenshots can include identifying details outside the passage an employee wants analyzed. Spreadsheets can contain hidden columns that remain in the uploaded file. Automated connections can send meeting transcripts to an AI service before anyone has reviewed or redacted them.
Sawant tackles that problem by first investigating the error himself, then deciding whether he can describe it safely to an outside AI service.
“I start by reproducing the problem in a controlled development environment, examining the error details, and isolating the component causing trouble,” he said. “If passing the problem to an outside AI service requires too much sensitive context, I continue troubleshooting with internal resources rather than sharing that information.”
Even after taking these precautions, an employee may still need to complete the task without AI.
Approved for what?
Company-provided tools appear to offer a way out of the trade-off. Employees can retain the context, work within an approved environment, and get on with the task.
But Sawant still checks the boundary. “An approved AI tool isn’t a free pass to share every kind of information,” he said. “You still need to understand what data it is approved to handle, how your inputs are stored or used, and who can access the conversation.”
Gage argues that employers should resolve the underlying questions before employees encounter them. Vendor contracts, data-processing terms, model-training arrangements, access controls, and retention settings belong in an organizational review.
Employees should receive the result in a usable form: which account to use, which tasks are permitted, and what information can go into it. “There are ways to use AI with personal and confidential business information, provided the appropriate contracts and safeguards are in place,” Gage said.
Meanwhile, employees face consequences for choosing the wrong route — even when they believe the information itself is harmless.
The wrong tool can cost a job
Howard Krauskopf, a Chicago litigation attorney who fields legal questions of this kind on AskaLawyerOnCall and JustAnswer, encountered that distinction when a mental health therapist sought advice. She was behind on billing and used ChatGPT to create a spreadsheet framework — even though her employer had directed staff to use a different, approved AI tool.
According to Krauskopf, the therapist told him she had entered no confidential client information. Krauskopf helped her explain to her employer what she had done and acknowledge the mistake.
Nevertheless, she was dismissed.
The account illustrates a distinction employees can easily overlook: avoiding a disclosure and complying with workplace instructions are separate obligations. More fundamentally, Krauskopf cautions, an employer generally does not need to establish misconduct to dismiss an at-will employee in the US, subject to legal restrictions and any applicable contractual protections.
“Ask before you use it, and put the question in writing,” he said. “If something goes wrong, you can show what you asked, what instructions you received, and why you believed you were following the rules.”
Documenting instructions may help employees persuade an employer to reconsider, even when it provides no guarantee of legal protection. How they present that record also matters.
Krauskopf advises a calm explanation that emphasizes the employee’s good record and desire to remain with the company. “Don’t attack,” he said. “Explain: ‘I followed the company’s instructions. Here is what I did. I like working here, and I’m asking you to take that into consideration.’”
In his experience, larger organizations are more likely to have formal procedures and management layers through which employees can seek reconsideration. At smaller companies, the outcome may depend more directly on an individual owner or manager. Neither setting guarantees a sympathetic hearing.
That leaves employees needing more than a warning about what could go wrong. They need guidance and safeguards that help them make the right decision before a problem arises.
How much judgment should the employee carry?
For Chris Kennedy, chief technology and security officer at insurance group Group 1001, a written policy alone is insufficient. “How many people go hunting for a policy, rather than asking, ‘What do I do in this situation?’” he asked.
“I never expect a policy statement to do all the thinking for the employee,” he said. “We try to write principles and then use examples and training to show how those principles apply.”
Those principles need technical support. At Group 1001, data classification and access controls help determine who can use particular information, while data leakage protection helps control where it can go. Kennedy describes an ongoing effort to consolidate data and label sensitive information so those controls can recognize and protect it. Monitoring and incident response help the company detect and address problems.
How much judgment remains with the employee depends on the job. Where tasks are predictable, Kennedy wants the company to establish the boundaries in advance. “For an operational role with a narrowly defined task, we want to set up the tools and controls so the employee doesn’t have to decide what data can go where,” he said.
Developers and teams exploring new products or investment strategies face less predictable decisions. As their work evolves, they may create proprietary information whose value — and need for protection — was not apparent at the outset. They need continuing access to managers and governance specialists who can help determine how that information may be used with AI.
Even routine work can change when existing software gains AI features. An employee may encounter a new assistant or automated function without deliberately signing up for another service. “What do you do when your existing third-party technology suddenly shows up with new AI-enabled functionality?” Kennedy asked.
Companies need to explain permitted uses, monitor how employees actually use AI, and update guidance and controls as new features and working practices emerge.
Making the safe choice workable
For Sawant, the support surrounding the rules determines how useful they are. “Getting help can be tricky because these questions come up right in the middle of solving a problem,” he said. “What helps is clear guidance with examples from the work I do, and someone in security or data governance I can go to for difficult cases.”
Managers also need to accept that a cautious decision can take time. “Employees should feel comfortable saying a task is taking longer because they’re waiting for guidance or approval,” Sawant said. “We need practical support so people don’t feel stuck or worried about making a mistake.”
Related reading:
AI policies work better when employees help write them
Surviving AI: Navigating workload creep, AI slop, and the new tech career playbookComputerworldRead More