Storm Clouds Over AI
A few days ago, my local daily paper, The Toronto Star, published an article about an overflowing toilet that left a couple with a $42,000 bill from the condo. The story didn’t spark an uproar. It was the photo. A reader contacted the paper and said she “was disappointed to see an AI-generated stock photo”.
Turns out the reader was right, and the Star published a full article explaining how it happened and that it contravened the Star’s policy on AI-generated content. The issue was that both the paper and the stock photo service missed it.
It was a small story about an overflowing toilet, but it’s a perfect snapshot of where we are with generative AI. That image was impressive enough to fool the pros and messy enough to prompt a newspaper to apologize. It didn’t fool a reader who noticed. It wasn’t supposed to go this way when ChatGPT arrived on November 30, 2022. Since then, lawsuits have been flying. The term “Deepfake” has entered the lexicon, and creatives are having awkward discussions with their clients when asked, “Is this image generated?”
The slippery slope to disillusionment
Like every new technology, generative AI has been riding the Gartner Hype Cycle. In 2024, Gartner had it sliding off the Peak of Inflated Expectations. By July 2025, it had landed in the Trough of Disillusionment. Gartner’s report, The Latest Hype Cycle for Artificial Intelligence, explains why: organizations spent an average of $1.9 million on generative AI initiatives in 2024, yet fewer than 30% of AI leaders said their CEOs were happy with the return. A year later, signs of recovery are emerging. Gartner’s 2026 update notes that some generative AI tools, such as virtual assistants, are moving toward the Plateau of Productivity.
Meanwhile, the hype has simply moved on. In April 2026, Gartner placed agentic AI at the Peak of Inflated Expectations. The cycle has started all over again. If you want to see that cycle up close, look no further than the launch of GPT-5 in August 2025. The buildup was enormous. OpenAI’s Sam Altman described it as having “PhD-level” expertise and a significant step on the path to Artificial General Intelligence. Then it landed with a thud. Users complained it felt colder and less capable than the model it replaced, and they were angry that OpenAI had removed the older GPT-4o without warning. The backlash was so loud that GPT-4o was back within about a day, and Altman admitted, “I think we totally screwed up some things on the rollout.”
The lesson here is not about OpenAI. It’s about us. Users didn’t judge GPT-5 on the launch event or the benchmarks. They judged it on whether it helped them do their work. That is the only measure that ever matters.
The Storm Clouds Gather
At the same time, the question of who owns what took hold. The Foundation models took a “Scrape now and ask for forgiveness” approach to building their models. This is when the lawsuits started flying. Bartz vs. Anthropic was one that I am involved in because 8 of my books were used to train Claude. One technique Anthropic used was to combine pirated books with books it purchased. This resulted in a $1.5 billion settlement, and cheques to the authors and publishers should be issued later this year.
Publishers also got involved, and one of the most visible cases is The New York Times vs OpenAI and Microsoft. The case was filed in 2023, and the judge is deciding whether it goes to trial.
One resolved case was Getty vs. Stability AI, which was heard in the UK. Getty lost because it claimed the Stable Diffusion model was an “infringing copy” of its images. The court disagreed, finding that the model learns patterns and doesn’t store or reproduce the photos.
For creatives, the rules on training models are still unsettled, which is why your tool choice matters.
Yes, but is it real?
The Toronto Star’s toilet is a small example of a much bigger problem. Scroll through any social media feed, and you will find AI slop: strange images, fake news stories and clickbait built to make you share it before you think about it. The question “Is it real or is it AI?” has become part of our daily routine.
It gets more personal with voices and faces. In 2024, Scarlett Johansson hired lawyers after OpenAI released a ChatGPT voice called “Sky” that sounded remarkably like her. She had turned down OpenAI’s request to use her voice, and OpenAI quickly pulled it.OpenAI insisted the voice belonged to a different actress, but it quickly paused “Sky” anyway. If a Hollywood star has to fight to protect her own voice, what chance do the rest of us have?
The darker side arrived in 2023 in Almendralejo, a small town in Spain, where AI-generated nude images of local schoolgirls started circulating among students. The case shocked the country and showed how easily these tools could be used to hurt people who never agreed to any of it.
The law is slowly catching up. In May 2025, the U.S. passed the TAKE IT DOWN Act, which makes it a crime to publish non-consensual intimate images, including AI deepfakes, and requires platforms to remove them within 48 hours. It’s a start. It won’t be the last.
Legislating the invisible. Rejecting the visible.
The companies building these models are not sitting still. Many of them are lobbying for a single federal rulebook that is light enough not to slow them down and strong enough to override tougher state laws. Governments are tentatively starting to respond. The EU now requires AI-generated content to be labelled. South Korea and China have similar rules, and Canada has just finished consulting Canadians on AI transparency, including how to label deepfakes. The U.S., meanwhile, is fighting its own states over who gets to write the rules.
Get Tom Green’s stories in your inbox
Join Medium for free to get updates from this writer.
Governments may be slow to act, but their citizens are not. And the fight is moving to the one part of AI you can actually see: the data centers. Governments may be slow to respond. Their citizens are not. The confrontation is shifting to the visible domain of artificial intelligence: the data center down the road.
Data centers have become a hot-button issue in communities across North America and Europe. That matters more than it might seem. Every AI model depends on massive amounts of computing power, and that power lives in data centers. If local resistance slows their construction, it slows the AI companies too. It may turn out that the first real brake on AI won’t come from Washington, Ottawa or Brussels. It will come from a town council meeting.
There is also growing nervousness about the enormous sums being poured into AI companies and data centres. Big tech alone is expected to spend more than $800 billion on AI this year, and both the International Monetary Fund and the Bank of England have warned that the AI boom looks a lot like a bubble. If it bursts, it won’t just be investors who feel it. AI spending has become one of the biggest drivers of U.S. economic growth. When the U.S. sneezes, Canada catches a cold, and the rest of the world reaches for the Kleenex. But history offers some comfort. The dot-com crash wiped out trillions, yet the fibre-optic cables laid during the boom became the backbone of the internet we use today.
Even so, there is a question to ponder if the bubble bursts: Are the AI companies too big to fail? Sam Altman, OpenAI’s CEO, appeared on the Conversations With Tyler podcast, where host Tyler Cowan observed that when things get huge, “the federal government is kind of the insurer of last resort, as we’ve seen in various financial crises and insurance companies screwing things up.” Sam’s reply? “I guess, given the magnitude of what I expect AI economic impact to look like, I do think the government ends up as the insurer of last resort, but I think I mean that in a different way than you mean that, and I don’t expect them to actually be writing the policies in the way that maybe they do for nuclear.” In other words, if the bubble bursts, the bill may end up with taxpayers. That’s all of us.
It isn’t all doom and gloom. On Gartner’s curve, the Trough of Disillusionment is followed by the Slope of Enlightenment. The hype burns off, leaving only the uses that genuinely work. We saw this with the web after the dot-com crash. The people who rushed in for the gold rush disappeared, while those who understood the craft stayed and built the web we use today.
The same will likely be true of generative AI. The trough is when the tourists leave. The professionals who learned to use these tools with skill and care will be the ones still standing. Maybe Hunter Thompson had it right when he wrote, “When the going gets weird, the weird turn pro.” You have to admit the AI world is weird. The good news is that the pros are the ones who stay.
Generate Responsibly
Three years ago, I wrote an article for the UX Collective calling for the ethical use of Generative AI. Since then, I have developed a few ideas on generating responsibly. They include:
Know where your tools come from.
For client work, choose tools that are transparent about their training data. Adobe Firefly built its reputation on licensed content, but be aware that it now also offers partner models, such as Google’s, that don’t carry the same promise. Check which model you are actually using.
Trust but verify.
Question every fact, every figure, and every image. AI is wrong more often than you think. The Toronto Star learned that the hard way.
Tell your clients.
Disclose AI use up front, starting with the proposal stage. This avoids the client asking: “Is this image generated?” Being upfront is a sign of professionalism, not an admission of anything. In fact, your clients may already have a policy on AI use. Ask if they have one.
Leave real people alone.
Don’t create anything that imitates a real person’s face, voice or likeness without their consent. Beyond the ethics, the law is catching up quickly.
Keep a record.
Save your prompts, iterations, and decisions. If questions arise about ownership or originality, your process is proof that the creative direction was yours.
None of this is complicated. It’s simply what professionals have always done: to take responsibility for their work
Your Takeaway
In Creative AI (Apress), which I co-wrote with Scott Fegette, we included a simple warning label: Generate Responsibly. Those two words sum up everything in this article. The lawsuits, the deepfakes, the bubble talk, and that AI-generated toilet are all real storm clouds. None of them is a reason to stop using these tools. They are a reason to use them responsibly.
The storm is real. Bring an umbrella, not an excuse.
