Forget the robot apocalypse. The AI threat that should actually worry you is already draining bank accounts and harvesting your personal data right now. Karolis Kaciulis, Lead System Engineer at Surfshark, revealed this week that the wave of dramatic extinction warnings circulating in tech circles is a recycled marketing tactic, not a genuine safety debate. And when you look at what AI is already doing to ordinary people, he has a strong point.
Speaking to TechRadar, Kaciulis argued that AI companies pull out the “rogue AI” narrative on a regular loop. The goal is to keep public attention locked on hypothetical Skynet scenarios while the actual, measurable harm gets ignored. It’s a neat trick. Doomsday headlines generate clicks, create a sense that only the companies building these systems are smart enough to control them, and neatly distract from questions about data privacy, financial fraud, and environmental cost.
The financial damage is not theoretical. Surfshark’s own research puts global losses from deepfake fraud at roughly $2.19 billion, with the UK alone absorbing $149 million of that. These are not future projections. This is money already gone, stolen through AI-generated voice scams, fake video calls, and automated fraud that bad actors can now run at industrial scale. Generative AI did not plot to conquer humanity. It just handed criminals a cheaper, faster, more convincing toolkit.
The environmental cost is also getting harder to ignore. A single ChatGPT query uses around 2Wh of energy on average, enough to run a 40W desk fan for three minutes. That sounds trivial until you multiply it across hundreds of millions of daily searches. The result is massive strain on power grids, land cleared for data centers, and enormous water consumption to keep server hardware cool. None of that makes the extinction debate, but all of it is real and growing.
Kaciulis also raises a point that rarely gets serious coverage: current large language models may be hitting a performance ceiling. If genuine scaling improvements are no longer possible, stoking fears about all-powerful superintelligence is a convenient way to mask a much more boring reality where the technology is simply plateauing. And a system that sounds confident and human does not equal one that is actually improving.
That illusion creates a direct privacy problem. Chatbots are designed to feel like helpful research partners. They ask follow-up questions, they sound natural, and people get comfortable fast. So they paste in private code, personal emails, and confidential documents without stopping to think that a corporate server is logging every word. The friendly interface is not an accident. It is the product.
So while tech executives warn about machines gaining consciousness, the more practical questions are: who owns the data you typed into that chatbot last Tuesday, and what happens to it? Those questions deserve more attention than the movie script Silicon Valley keeps selling.