Fired for secret AI use? Employers who set no rules share the blame
Millions of workers use AI in the shadows because nobody told them how. Dismissal can be legitimate, but a company with no policy has written half the script.
Photo: Bill Branson (Photographer) / Wikimedia Commons (Public domain)
Is it fair to fire someone for using a tool their employer never forbade, never provided and never even mentioned? Legally, the answer may be "yes, sometimes." Morally and practically, we think the answer is "not before the company looks in the mirror."
The question is on the table because Der Spiegel asked labor lawyer Anton Barrein when secret AI use can cost you your job. The framing is blunt: millions of employees use AI, many of them in secret, and a dismissal can follow in some circumstances. The piece also covers what companies and employees should settle so it never gets that far. We only have the teaser, so we won't pretend to know which cases Barrein calls risky. The direction of the argument is clear enough.
Secrecy is a symptom, not a character flaw
If hiding AI use were a fringe habit, "bad employee" would be a fair diagnosis. It isn't a fringe habit. A Cybernews survey found that nearly 6 in 10 US employees use unapproved AI at work, often exposing sensitive corporate data. A PagerDuty report describes the same pattern from the other side: employee AI adoption has outpaced corporate policy at large organizations.
When that many people behave the same way, the cause is rarely individual moral failure. It's probably a vacuum. Workers see a tool that can save them time, they see no approved alternative and no guidance, and they do the rational thing: quietly use it. Hiding it is also rational, because nobody knows whether disclosure will be rewarded or punished. Another survey reports that 83% of employees fear being fired for an AI mistake. That figure comes from a survey of 2,500 global employees and IT leaders, and we only have its summary. Even so, fear like that doesn't produce transparency. It produces silence.
Yes, the risk is real, and employees shouldn't shrug it off
None of this makes shadow AI harmless. Pasting customer data, contracts or source code into a consumer chatbot is exactly the kind of thing that justifies a serious conversation, and in some cases more. The same Cybernews finding mentions sensitive corporate data being exposed. That is the legitimate core of the employer's case, and it's why dismissal can plausibly be a real possibility.
So our advice to workers is simple, and it isn't "stop using AI." Don't put confidential material into tools nobody approved. Ask in writing what's allowed, even if the answer is "we don't know yet." And be honest about the difference between "I used a tool to polish an email" and "I fed it a client file." Employers and courts will likely not treat those two the same, though that's our reading, not a legal ruling.
Blame is shared, and some bosses are the worst offenders
Here's the part we find most telling. A CIO.com piece on a BlackFog survey says roughly half of employees use unsanctioned AI tools, and its headline names enterprise leaders as major culprits. If the people setting the rules are also bending them, any dismissal for the same behavior starts to look selective at best.
Then there's the opposite failure: companies that demand AI use while giving no framework. One tracker of workplace AI mandates notes that such requirements can be enforced through performance reviews, promotion criteria, headcount approvals or even firing for non-adoption. Think about what that does to an employee. Use AI and you may be punished. Don't use it and you may be punished. Use it openly with no policy, and you're gambling on your manager's mood.
Even the companies pushing hardest can stumble on communication. Proton, as Golem reports, scrapped its re-interviews on AI use, and Yen admitted a communication error while saying the company will keep pushing AI. We'd take that as a small lesson: the plan can be fine and the rollout still poisonous.
Bans are the lazy answer; a decent policy is boring and concrete
Bans are the lazy answer, and in our view a losing one. If half your staff already uses these tools, a ban doesn't remove the behavior. It removes your visibility into it. What works is boring and concrete:
- A sanctioned tool that's actually good enough that people don't need a personal account.
- A short, plain-language list of what data never goes in.
- A named person to ask, and a promise that asking is safe.
- Training that's tied to real tasks, not a once-a-year slideshow.
None of that is likely to be expensive compared with a data leak or a lawsuit over a questionable dismissal.
Our verdict
If you're an employee, treat undisclosed AI use with confidential data as a genuine career risk, and get your employer's position in writing. If you're a manager, understand that every "secret" user is probably evidence of a gap you created. Firing someone may be legal. Firing someone for filling a gap you left open is, at the very least, a confession that you never did your job.
No topics yet: start the first one.
More stories
AI's "zero payoff" isn't a model problem. It's how companies work and count
Apollo's economist says AI hasn't paid off outside tech. We think the gap lies less in the models than in unredesigned work and in how companies measure results.
Fifty AI agents built a bureaucracy, not a civilization, and that's the lesson
A Minecraft experiment where 50-agent teams drowned in paperwork is a useful cold shower: adding agents mostly adds coordination cost, not intelligence.
A rumored Mythos leak matters even if it's fake: weights need key-grade security
The alleged Anthropic checkpoint on Hugging Face is unverified. The real lesson, in our view, is that labs should guard their weights like cryptographic keys, not like ordinary company files.
Brazil's prompt injection fine is right, but fines alone won't protect courts
Brazil's top appeals court fined lawyers for hiding commands aimed at its AI. Sanctions are a fine first move, but every document a machine reads is now an attack surface.
