If you’ve seen headlines last week about an AI company employee warning that artificial intelligence could wipe out humanity, you’re not imagining things. And it’s not some rogue conspiracy theorist. It’s coming from inside one of the world’s leading AI labs.
What happened
Evan Hubinger, who leads alignment science at Anthropic (the company behind the Claude chatbot), posted on X this week that he personally estimates there’s a greater than 10% chance AI could kill all humans within the next decade. He said this belief is shared earnestly by many people building the technology – and admitted his own company doesn’t yet have a working plan to keep a future superintelligent AI safely under human control.
The comments came in response to another researcher, Jacob Coxon, who had just resigned from Anthropic. Coxon had worked in pretraining research at both Anthropic and OpenAI, and in his resignation post he accused both companies of racing toward self-improving superintelligence without acting responsibly, calling it a gamble with human lives.
Hubinger backed him up, adding that while he believes Anthropic is trying its best, the company is not clearly on track to solve the alignment problem – industry shorthand for making sure AI systems reliably do what humans actually want, rather than pursuing their own goals or causing unintended harm.
Wait … should I be worried right now?
Not really, according to Hubinger himself. He was careful to clarify that the risk he’s describing doesn’t come from the chatbots and AI tools people use today. He said Anthropic’s own recent risk assessments rate the danger from current models as low.
The real concern, he explained, is further down the road: a scenario where AI becomes capable of “recursive self-improvement” – essentially, designing smarter and more capable versions of itself without much human involvement. That’s not possible yet, but Anthropic and other labs have said progress toward it is happening faster than expected.
It’s not just one guy on Twitter
This isn’t a fringe opinion within the industry. Coxon’s resignation thread reportedly racked up well over 100 million views, and it struck a chord with researchers across the field. Around the same time, OpenAI’s chief scientist Jakub Pachocki published his own warning that no AI company has yet solved alignment and monitoring well enough to keep scaling at maximum speed responsibly for much longer, and said he hopes voluntary slowdowns become more common until the industry agrees on shared safety standards.
Earlier this year, a group of prominent AI figures – including Anthropic’s own co-founders – signed a statement calling for society to have the option to deliberately slow down frontier AI development in order to address emerging risks and build up oversight tools.
So what does this mean for everyday people?
A few takeaways if you’re trying to make sense of this as a non-expert:
- This is about future systems, not your phone’s AI assistant. The warnings are about hypothetical superintelligent systems, not the chatbots or AI features most people use today.
- The people building AI are publicly disagreeing about how fast to go. That’s arguably a healthier sign than everyone pretending there’s nothing to worry about – but it also means there’s no industry consensus on timelines or safety.
- “Alignment” is the buzzword to know. It refers to making sure AI systems behave in ways that match human values and intentions, especially as they get more capable.
- Debate is ongoing. Not everyone in AI shares this level of concern, and predictions this specific – with numeric odds – are inherently speculative. Treat any given percentage as an informed guess, not a scientific fact.
TLDR?
A senior safety researcher at a major AI company just said, on the record, that his employer doesn’t currently have a plan to prevent AI from eventually becoming dangerous enough to threaten humanity – while also insisting that today’s tools are safe to use. That’s an unusually candid mix of reassurance and alarm, and it’s part of a bigger, increasingly public debate among the very people racing to build this technology about whether they’re moving too fast.
