Deepfake fraud has already cost the world roughly $2.19 billion. That number is real, documented, and happening now. So why is the AI conversation still dominated by hypothetical extinction scenarios? According to Android Headlines, Karolis Kaciulis, Lead System Engineer at Surfshark, argues that the doomsday rhetoric flooding tech news is a calculated distraction, not a genuine safety debate.
Kaciulis spoke to TechRadar to make a pointed case: AI companies recycle the same “rogue AI” narrative on a regular loop. Each wave of apocalyptic headlines conveniently redirects public attention away from the immediate consumer damage these tools are already causing. It’s a pattern that repeats whenever a major release needs hype or a data scandal needs burying.
The actual harm isn’t a Skynet scenario. It’s generative AI giving bad actors a cheap, scalable toolkit for automated scams, voice cloning, and digital fraud. The $149 million lost to deepfake fraud in the UK alone is a concrete signal of where the real risk sits. And that’s before you factor in the environmental cost. A single ChatGPT query consumes around 2Wh of energy on average, enough to run a 40W desk fan for three minutes. Multiply that across hundreds of millions of daily queries and the pressure on power grids, land use for data centers, and water consumption for server cooling becomes genuinely significant.
There’s also a more cynical reason companies might prefer you focus on sci-fi futures rather than present-day limitations. Kaciulis suggests that large language models may be hitting a performance ceiling. Promoting existential danger is a useful way to mask the fact that meaningful scaling gains are harder to come by with current architectures. Instead of breakthroughs, systems may mostly be getting better at producing the responses users already expect, which looks like progress without necessarily being any.
That illusion creates a direct privacy risk. Chatbots are designed to feel conversational and helpful, and that comfort makes people careless. Users routinely paste private code, personal emails, and confidential documents into text boxes without considering that a corporate server is logging everything. The friendly interface is doing real work here, and not in the user’s favor.
This critique fits into a broader industry trend. OpenAI, Google DeepMind, and Anthropic have all published or signed statements warning about catastrophic AI risk, often timed around funding rounds or regulatory hearings. The pattern is hard to ignore. And with AI regulation accelerating in the EU and the US, the framing of “existential threat” is also politically useful, steering lawmakers toward big-picture science fiction rather than the proximate harms that existing consumer protection law could actually address today.
The practical takeaway is straightforward. Reviewing privacy settings, avoiding sharing sensitive files with chatbots, and learning to spot deepfake voice scams are more useful right now than debating machine consciousness. The threat model that actually affects most people is already here. It just doesn’t make for a good movie trailer.
