So Samsung banned ChatGPT for its own employees. I read about this a few days ago and haven't stopped thinking about it since, mostly because I almost did the exact same dumb thing that got them into trouble.
Quick version if you missed it: some Samsung engineers were pasting internal stuff into ChatGPT to get help: one guy reportedly dropped in actual proprietary source code to get it to find a bug, someone else pasted meeting notes to get a nice summary out the other end. Three separate incidents inside about three weeks, according to the reporting. And the thing people keep forgetting about ChatGPT (at least the free/consumer version) is that whatever you type into that box doesn't just vanish into the void. It can get used to train the model further. Which means, in theory, a chunk of Samsung's semiconductor source code is now sitting somewhere inside OpenAI's training pipeline, or at least it might be, and nobody at Samsung can un-ring that bell. So the company's response was blunt: ChatGPT and tools like it are now off-limits on company devices and networks, and if you use your own phone for it you're capped on how much you can upload.
I get why this is happening at basically every company with a legal team right now, not just Samsung.
Here's my confession. A couple months back I was debugging a genuinely annoying async bug in some client code (not proprietary in the "worth billions" sense, just a normal freelance NDA situation), and I had the file open, and I almost just pasted the whole thing into ChatGPT to ask "why is this racing." I stopped myself about two seconds before hitting enter, mostly because I remembered the NDA had a specific clause about not sharing code with third-party services, and legally a chatbot definitely counts as a third party. I ended up retyping a stripped-down, renamed version of the function instead, which took longer and honestly worked fine, but it bugged me the whole time that I was doing extra work just to be careful with a tool that's supposedly here to save time.
That's the part nobody's really talking about enough. Everyone's either in the "AI is going to write all our code by 2025" camp or the "its just autocomplete, ignore it" camp, and almost nobody's dealing with the boring middle problem, which is: this is a genuinely useful tool that a huge number of white-collar workers now have access to, and basically none of the corporate policy or legal infrastructure exists yet to say what's okay to feed into it. IT departments are stuck writing memos in real time. I'd bet money there's a Slack channel at some large company right now where someone asked "can I paste the Q2 roadmap into ChatGPT to make it sound more professional" and got three conflicting answers from three different people, none of whom actually knew the policy because there wasn't one yet.
The other thing that gets me is how fast this went from "cool toy I show my parents" to "actual liability the legal department has opinions about." That's maybe a five month turnaround, from ChatGPT launching at the end of November to companies drafting bans. I don't think I've seen a piece of consumer software cross that particular threshold (hobbyist curiosity to compliance memo) that quickly before, not even the iPhone, and that thing changed everything.
Practical note if you're a freelancer like me and don't have an IT department to tell you what to do: I've started keeping a mental (now actual, it's a sticky note on my monitor) rule that anything client-related gets either genericized before it touches any AI tool, or doesn't touch one at all. It's a small hassle. It's also just obviously the right call once you think about it for more than ten seconds, and I'm a little annoyed at myself that it took nearly hitting enter on that debugging session to actually make the rule instead of just having good instincts from the start.
Anyway. My personal wifi router has been acting up all week too, dropping connection every few hours for no reason I can find, completely unrelated to any of this, and it's been driving me up the wall more than a hypothetical AI data leak ever could. Priorities.