Your Company Blocked ChatGPT. Now What?

I talk to a lot of people in this exact spot. ChatGPT and Claude are blocked on the work network. You’ve got a Microsoft Copilot license nobody explained properly, or an internal chatbot the company bought a platform for and half-built. You sat through an intranet course on “ethical AI use,” maybe a second one on basic prompting, and somewhere in the building there’s a person or two whose job title has “AI” in it and whose actual mandate nobody, including possibly them, could describe to you.

Most AI advice assumes you can just pick a tool and go. Open Claude, make an account, start using it for real work. That’s fine advice for a freelancer or a founder. It’s useless for most employed people, because most employed people don’t get to choose.

If that’s you, this is the guide. It won’t try to sell you on AI; you’ve probably already tried it privately and know what it can do. This is about what to actually do when the good tools are off-limits and the sanctioned one feels like a consolation prize.

Why they blocked it — and why that’s not stupid

Start with the uncomfortable truth: your IT department is usually right to block the free consumer versions of ChatGPT and Claude for work use.

A free ChatGPT account isn’t built for handling a client’s contract, a patient record, or your company’s unreleased pricing. Depending on the plan and settings, your inputs can be used to improve the model, retention policies aren’t what a regulated business needs, and there’s no data processing agreement behind it. If someone pastes a client email into their personal ChatGPT to draft a reply, that data has left the building in a way nobody at the company agreed to or can account for. For a company handling client data, health information, or anything under a confidentiality agreement, that’s a real liability, not a hypothetical one. Regulatory pressure only makes the case stronger — a compliance team that waves this through is doing their job badly.

So the block itself is a reasonable decision. Where it goes wrong is what usually happens next. IT blocks the risky tools, procurement buys a sanctioned platform, someone runs a mandatory training, a policy document goes up on the intranet, and leadership quietly checks a box: AI, handled. Nobody comes back to ask whether people actually got good at using it, or whether the training taught anything beyond “don’t paste client names into random websites.” The topic gets filed next to expense reports and data protection modules — something you comply with, not something you get better at. That’s the real failure, and it’s on the company, not on you.

Look at what you actually got before writing it off

Before you dismiss Copilot or the internal chatbot as the consolation prize, check what’s actually running underneath it.

Copilot inside Microsoft 365 runs a frontier-class model. My complaints are about the packaging, not the model: it’s not as configurable as ChatGPT, it doesn’t learn about you as cleverly, and it interacts with other software surprisingly poorly, amazingly even with Microsoft’s own products. But “the AI is dumb” isn’t on that list. Your internal chatbot might be running something similarly capable underneath a narrower interface.

The part that’s easy to miss: the tool you’re allowed to use on real work beats the better tool you’re not allowed to use on real work. Your personal free ChatGPT account is genuinely more polished in places. But you can’t safely put a client name, a real contract, or an internal number into it. Copilot or the internal chatbot exists specifically so you can. That’s the entire point.

So find out what you’re actually permitted to do. Read the policy if one exists. Ask whoever administers the tool what kinds of data it’s cleared for. Most people never do this and just assume the sanctioned tool is useless because it feels unfamiliar, when the real gap is that nobody told them what it’s for.

The two-track approach

The practical move: run two tracks, and let the skills carry over, not the accounts.

Track one, private. Use a free account on your own device, with material that’s genuinely yours: planning a trip, drafting a cover letter, summarizing a rental agreement, working through a hobby project. This is where you build the actual skill — giving the AI a real goal and the relevant context, iterating instead of expecting a perfect answer on the first try, asking it to interview you when you’re not sure what it needs to know, checking its output instead of trusting it blindly. None of that depends on which tool you’re using. It’s the same skill whether you’re talking to Claude on your phone at night or Copilot in Outlook at 9am.

Track two, work. Take those same habits into whatever your company sanctioned. Same briefing discipline, same follow-up questions, same skepticism about the first answer. The tool might be clunkier and the model might be a step behind what you tried at home. That’s fine — the skill is what does the heavy lifting, and it works whether or not the interface is nice.

The two tracks never need to touch. Your private account never sees work material, and your work account benefits from everything you learned privately. The learning compounds even though the data never crosses.

If you want the actual mechanics of that learning-by-doing approach — how to brief an AI properly, how to keep a conversation going instead of treating every message as a fresh query — I’ve written that up in Getting Started with AI and How to Actually Get Good at AI. Both apply directly to whichever tool you’re using on either track.

The one hard line

Don’t put real work data through your private account to get around a limitation in the sanctioned tool. Not the client’s email thread, not the internal doc, not the meeting notes with a colleague’s performance details in them — even if you’re just trying to get a better draft faster and you’d never actually leak anything on purpose.

I get why people do it. The sanctioned tool feels slower or dumber, and routing around it feels like initiative, not rule-breaking. But this is exactly how client data ends up somewhere it was never supposed to be, and exactly how people lose their jobs over something that felt like a shortcut at the time. If the tool your company gave you genuinely can’t do something you need, that’s useful information — tell whoever owns the AI rollout, in specific terms, what’s missing. That’s feedback that can change the tool. Quietly working around the policy just hides the gap and adds risk nobody agreed to carry.

What the intranet training actually missed

Those mandatory courses aren’t wasted time — knowing the rules and not embarrassing yourself with an unsafe prompt matters. But they teach compliance, not capability. Nobody tested whether the “basic prompting” course changed how you actually work.

The piece that’s usually missing is the mental model: think of the AI as a new colleague who is universally well-read but knows nothing about your company, your team, or the specific thing you’re working on this week. A new colleague needs a proper brief, not a one-line request, and gets better the more you talk it through with them rather than expecting a finished answer on message one. That shift matters more than any list of prompt formulas, and it’s exactly what a two-hour compliance-flavored training usually skips. Getting Started with AI covers it in detail if you want the fuller version.

Go find the AI champions

Somewhere in your company there’s probably one or two people whose job is loosely “find AI use cases” or “AI champion” or something equally vague. Most people either roll their eyes at this or ignore it entirely, assuming it’s a box-ticking exercise with no teeth.

Go talk to them anyway. Bring one specific, boring task from your own work — a report you write every month, a type of email you send too often, a document you always have to reformat. Most of these people have a mandate and very little support, and a concrete example from someone in the building is worth more to them than another workshop slide. You might end up being exactly the kind of grounded, visible use case that makes the rest of the rollout make sense to other people.

If you’re curious what your leadership is probably getting wrong about this whole thing — and what doing it well actually looks like — I wrote about that from the other side in Giving Everyone Copilot Isn’t an AI Strategy. It might explain a few things about why your company’s approach feels so hollow.

What to do Monday

Tonight, on your own account: pick one small, real, private task — an email you’ve been avoiding, a decision you’re stuck on, a document you need summarized — and work through it with a free assistant, actually iterating instead of accepting the first answer.

Tomorrow at work: take the same approach into Microsoft Copilot or whatever your company gave you, on one small real task where you already know what a good answer looks like.

And somewhere this week: ask whoever owns the AI policy what your sanctioned tool is actually cleared to handle. Most people never ask and just guess wrong in the cautious direction, which means they underuse a tool their company already paid for.

None of this requires permission you don’t have. It requires using what you’ve already got, on both tracks, on purpose. If you want a structured way to track where you actually stand, the AI Fluency Path turns this into a two-minute check and a level-appropriate next step, whether you’re just getting started or already past the basics.