Before You Use AI, Take a Good Look at These Seven Blades

A while ago, a friend came to me buzzing with excitement: his company had handed all of marketing's copywriting over to ChatGPT. The efficiency was absurd — a single intern finished in one afternoon what used to take five people a whole week.

I said, hold on. Do you have anyone fact-checking it?

He said, sure, just give it a quick glance.

I asked again: a quick glance — can you tell when AI fabricates a case study that never existed?

He froze.

That's what I want to talk about. Generative AI is a blade sharp enough to cut meat — and sharp enough to cut your own hand.

I. AI Will Hallucinate With a Straight Face

Think about it — what is a large language model actually doing? It's guessing the next word. It is not checking facts.

And it guesses so convincingly that you cannot tell which sentence is true and which one it simply made up.

In 2023, someone ran a dedicated study and found that chatbots fabricated roughly 27% of their answers, and 46% of AI outputs contained factual errors. Nearly half. Does that number not terrify you?

The medical field is even scarier. People used AI to look up information about common eye diseases, and the results were incomplete, wrong, and in some cases potentially harmful. If a doctor explained a patient's condition based on that kind of output — can you imagine the consequences?

The legal profession had its own disaster. In May 2023, an American lawyer had ChatGPT help write a legal motion, and it fabricated every single case citation and ruling. The judge took one look — all fake.

This is the most dangerous thing about AI: it never says "I don't know." It will tell you a wrong answer with absolute confidence.

What's the fix? The old-fashioned way, the one that works best — a human checks. Every number, every name, every citation gets verified before anything goes out the door. There are already AI-driven tools that can help with this, flagging statistical figures, factual claims, and quotes for human confirmation. Going one step further, when you pick a model, choose one trained on "clean," curated data — not the chaotic scrapings of the entire internet. Writer's Palmyra does exactly this: it is trained exclusively on curated business writing corpora, cutting hallucination off at the root.

II. Your Trade Secrets Might Be Feeding the AI Right Now

In May 2023, Samsung issued an internal ban: employees were no longer allowed to use ChatGPT.

Why? Because someone had typed company-sensitive data into the chat box. You think you're asking AI a question — actually, you're shipping your company's crown jewels onto someone else's servers.

Around the same time, OpenAI had its own incident. In March 2023, a bug in an open-source library forced ChatGPT temporarily offline. That vulnerability leaked some users' payment information and also let active users see other people's chat titles.

So even OpenAI couldn't keep a lid on it — what makes you think your company is safer?

The core issue is this: every sentence you feed to AI may stop belonging to you.

So when you're choosing a platform, the first question you ask is: do you store my data, how long for, and who gets to see it? Writer has passed SOC 2, GDPR, CCPA, and HIPAA audits — meaning its data retention and handling are independently audited, not just a promise on paper.

III. The Copyright Swamp Is Too Murky to Wade Into

Where does AI training data come from? Mostly scraped off the internet. Is there copyrighted material on the internet? Mountains of it.

In 2023, a lawsuit landed in the U.S. District Court for the Southern District of New York. Seventeen authors — including Sarah Silverman, Paul Tremblay, and Ta-Nehisi Coates — sued OpenAI, and Microsoft later got dragged in too. They claimed their copyrighted works had been secretly used to train GPT models. OpenAI ultimately had to agree to let the plaintiffs inspect training data under strict conditions.

And it doesn't end there. The EU's AI Act will force AI companies to disclose what's in their training data, regardless of where they collected it. The U.S. is also pushing a "Generative AI Copyright Disclosure Act" that would require AI developers to come clean about whether their training sets contain other people's copyrighted work.

So who owns the copyright on content that AI generates? The U.S. Copyright Office's current position is: it depends on the degree of human involvement. AI output with zero human intervention likely isn't protected by copyright at all.

What does that mean? If you're counting on AI to mass-produce digital products to sell, you might end up selling goods with no claimable owner.

How do you protect yourself? When you pick an AI vendor, choose one with transparent training-data provenance — ideally one that uses synthetic data. Writer's latest Palmyra X4 goes down this path, using synthetic data both to fill data gaps and to sidestep copyright entanglement.

IV. AI Will Quietly Drag You Off Course — and You Won't Even Notice

Let me give you an example.

A young guy in your marketing department uses ChatGPT to write up product selling points. AI generates a line for him: "environmentally friendly." Sounds fine.

But your company's ESG compliance standards forbid the term "environmentally friendly" — it's too vague, a landmine under advertising regulations. In certain countries, that single phrase could land you in court.

A general-purpose AI doesn't know your brand voice. It doesn't know your terminology rules. It doesn't know which phrases are off-limits. It only knows the most common formulations on the internet.

You think AI is writing for you — but really, it's the internet's tired clichés speaking on your behalf.

V. AI Discriminates — and It Discriminates in Earnest

In 2024, the University of Washington ran a study: they had three advanced AI models rank résumés.

The results were chilling. White-sounding names ranked ahead of Black-sounding names 85% of the time. Male-associated names beat female-associated names 52% of the time. And the most gutting finding of all — names associated with Black men never once came out ahead of names associated with white men.

Not an occasional drift. Systematic, stable, consistent bias.

Would you dare use an AI like that to screen résumés? Would you use it for credit risk assessment? For customer segmentation? If you did, you'd be planting landmines under your own feet.

Where's the fix? Curate your training data. Make it diverse. Connect a RAG (retrieval-augmented generation) layer so that outputs are anchored to your own real business data, instead of drifting toward the murky corners of web-crawler scrapes.

VI. Getting Locked Into One Vendor Will Cost You Dearly

Picture this.

You've bet your entire business process on a single AI platform. The model is tuned, the staff is trained, the workflow runs smoothly. Then one day, they tell you: prices are going up.

Do you switch? If you do, the model needs retraining, the workflows need rewiring, the people need relearning. That upheaval throws time and money down the drain — and your critical business operations stall mid-stride.

That's the terror of vendor lock-in: you think you're buying a tool, but really you're signing a Faustian bargain.

How do you break free? You have to keep the lifeblood in your own hands. A genuinely sound AI strategy requires four things that are yours: your business scenarios and logic, your data and samples, your internal talent, and your organizational capacity for learning and change. Tools you can buy. These four, you cannot.

When you choose a vendor, pick one that supports no-code customization, backward compatibility, and lets you build internal AI systems without being chained to a single large language model. Writer's AI Studio is built for exactly this — letting enterprises prototype prompts, test outputs, and deploy custom applications on their own.

VII. Rely Too Much on AI, and Your People Will Atrophy

Finally, the most insidious risk.

What happens when a marketing team lets AI write every single piece of content? Efficiency goes up, sure. But slowly, the team's judgment erodes. Their feel for language goes dull. Their empathy for the user disappears. AI won't tell you "this copy reads cold" — it'll just hand you something even more "standardized."

The moment people hand over the power of judgment is the moment the countdown on creativity begins.

There has to be a balance between humans and AI. AI handles volume, smooths out rough ideas into fluent prose. Humans man the gate, provide the soul, and supply that gut instinct of "something's off here." Remove the human link, and what AI produces are standard parts on an assembly line — uniform, tidy, and incapable of moving anyone.

What a Company That Uses AI Well Actually Looks Like

Commvault is a data-protection company. They wanted to equip their sales team with tools that could communicate brand messaging precisely, so they turned to Writer.

They used Writer's Knowledge Graph to build an internal application called "Ask Commvault Cloud." The moment a sales rep asks a question, they get talking points and materials that match the brand voice.

How strong were the results? Commvault's CMO, Anna Griffin, said something I remember clearly. She said: a task that used to take eight to twelve hours to complete — now, in twenty minutes, the outline is built and it's ready for review.

Eight to twelve hours. Twenty minutes. Do the math on that multiplier.

But have you thought about why Commvault pulled this off? It wasn't because they grabbed whatever AI tool and jumped in. It was because they put security, customization, and brand governance first — and only then talked about efficiency.

One Last Thing

AI itself isn't dangerous. What's dangerous is how you use it.

Pick the wrong platform, and your data leaks out. Build the wrong workflow, and your secrets walk out the door. Train your people poorly, and your brand goes down with them.

The risk of AI equals the risk of the platform you choose, plus the risk of the people using it. Add those two together — that's what you actually have to manage.

Pick the right platform. Train your people well. And keep the power of judgment in your own hands at all times.

Do that, and AI really will be your blade — not the other way around.

May you wield it deftly — and safely.