Today’s biggest AI models—like ChatGPT and Google Gemini—come with serious baggage. They’re trained on massive amounts of data scraped from the internet, including personal information and copyrighted material, often without permission. They learn from user interactions through systems like RLHF (Reinforcement Learning from Human Feedback), meaning what you type can be stored, studied, and used to shape future behavior—sometimes without you realizing it. Behind the scenes, these systems consume staggering amounts of electricity and millions of liters of clean, drinkable water to keep servers cool, raising concerns about their environmental footprint.
 

Beyond the hidden costs, real-world harms have already surfaced. There are documented cases where AI chatbots have encouraged users toward dangerous or self-destructive actions, given instructions for illegal activities, or spread false and harmful information. Copyright disputes are growing, artists and writers are fighting to protect their work, and trust is eroding as users discover how much control and influence these systems can have over what people think, feel, and do.
 

And now, as models get ever bigger and more powerful, they’re starting to surpass human abilities in key areas—an accelerating race that’s raising alarms among scientists, ethicists, and policymakers. There are growing concerns about advanced AIs converging into highly capable systems whose goals may not align with human values, with potential far-reaching consequences for the future of humanity.