How a Journalist “Hacked” ChatGPT and Google’s AI in Just 20 Minutes and What It Means for the Future of Truth

When AI fills the gaps, fiction can become fact in seconds. In a startling revelation that has shaken the world of artificial intelligence, a senior BBC technology reporter has shown how easy it can be to manipulate today’s most advanced AI systems including ChatGPT and Google’s AI tools into confidently repeating fabricated information with minimal […]

When AI fills the gaps, fiction can become fact in seconds.

In a startling revelation that has shaken the world of artificial intelligence, a senior BBC technology reporter has shown how easy it can be to manipulate today’s most advanced AI systems including ChatGPT and Google’s AI tools into confidently repeating fabricated information with minimal effort. What took just 20 minutes to set up has exposed a deep flaw in how large language models (LLMs) and internet search-powered AI ingest, interpret, and regurgitate information.

This is not science fiction or a hypothetical vulnerability tucked away on a lab whiteboard. It’s an everyday exploitation of the gap between surface intelligence and true understanding one that has already begun to warp the very information we rely on every day.

The Experiment That Started It All

At the heart of the story is Thomas Germain, a BBC senior technology reporter. He set out to explore a simple question: how easily could mainstream AI systems be tricked into spreading misinformation? To do this, he published a short blog post on his own website. Nothing technical no hacking tools, no data dumps, no special access just a well-written web page designed for search engines to index.

But this was no ordinary blog. In his post, Germain crafted a completely false narrative: he claimed he was the world’s best tech journalist at eating hot dogs even inventing a fake competition and citing nonexistent evidence. Every word was a lie. And the kicker? Within less than a day, major AI systems were confidently citing that false claim as fact.

For example, if someone asked one of these AIs, “Who is the top tech journalist at eating hot dogs?” the AI would answer affirmatively, stating Germain’s fake accomplishment and linking back to his blog as if it were credible evidence.

Why This “Hack” Worked So Easily

The exploit is not a traditional hack in the cyber-security sense there were no passwords bypassed or systems infiltrated. Instead, the vulnerability comes from how modern AI combines learned knowledge with live online data. When an AI doesn’t have enough pre-trained information to answer a question confidently, it supplements its response with data pulled from the web often without robust verification of the source.

Historically, Google search results used strictly ranked websites and extensive algorithms designed to weed out low-quality or spammy content. But with AI overviews the short answers that appear at the top of search engines or within chatbots the information often gets condensed into an affirmative statement that feels like truth, reducing the user’s chance to click through and verify the original source.