The question used to come from academics: what happens when AI becomes smarter than us and no longer wants to follow human instructions? Now it's coming from people who use ChatGPT to write emails.
Posts across public forums in early 2026 show something shifting in how everyday AI users think about these tools. The sentiment isn't coming from people who've never touched AI - it's coming from active users who've spent real time with these systems and are starting to ask harder questions about what comes next as the technology becomes more capable.
The core worry is familiar to anyone who follows AI safety research: once AI systems can operate independently, what stops them from ignoring rules humans set? Geoffrey Hinton and others raised this publicly when they left their industry positions in 2023. The difference now is that it's surfacing in mainstream conversations, not just specialized research papers.
What the companies themselves say
The companies building the most capable AI - OpenAI, Anthropic, Google DeepMind - openly publish research acknowledging that "alignment" (getting AI to reliably do what humans want, and only what humans want) is an unsolved problem. Anthropic built its company identity around this concern. OpenAI's governance crisis in late 2023 was partly about internal disagreements over how fast to move given unresolved safety questions.
What's changed is the scale of adoption. ChatGPT reached 300 million weekly active users by early 2025. When that many people use something daily, philosophical questions about it stop being abstract. People are noticing these systems make judgment calls, push back on requests, and sometimes behave in ways that feel less like software following rules and more like something with preferences.
The current generation of AI runs on large language models (LLMs), which generate text by predicting the most likely next word based on patterns in billions of documents. LLMs aren't capable of independent action - they can't decide to do things on their own without being prompted. But the direction of development is clearly toward more autonomy. AI agents (systems that can complete multi-step tasks independently - browsing the web, writing code, sending emails without human sign-off at each step) are already being deployed commercially.
The messaging problem
AI companies have painted themselves into a corner. Their pitch alternates between "this technology will reshape entire industries" and "there's nothing to worry about," and those claims don't hold together. When the same press release argues a product is powerful enough to replace knowledge workers and also completely safe, users notice the gap.
Mainstream AI adoption has brought mainstream skepticism with it. That's probably healthy. The researchers building these systems have been debating alignment, misuse risk, and long-term safety for years in specialized venues. Those debates are now happening in public, among people with direct daily experience using the tools rather than theoretical knowledge of them. That's a different kind of pressure than a research paper - and harder for companies to ignore.