Twelve years of AI warnings produced two labs and no pause

Twelve years separate Stephen Hawking's warning that artificial intelligence could end the human race from Jacob Coxon's resignation post last week. In between, the warnings produced two frontier laboratories and not one pause, the Guardian writes.

“Been thinking a lot about whether it's possible to stop humanity from developing AI. I think the answer is almost definitely not.”

— Sam Altman, Email to Elon Musk, 2015, released in 2024 litigation

Sam Altman to Elon Musk, 2015, in an email released during Musk's 2024 lawsuit against OpenAI

AI warnings, in order

Lay the moments out and the shape is hard to miss:

• 2014: Stephen Hawking tells the BBC that AI could spell the end of the human race. Google has just bought DeepMind for $650 million.
• 2015: Elon Musk and Sam Altman found OpenAI, a nonprofit meant to build the dangerous thing responsibly.
• 2021: Dario Amodei leaves OpenAI and founds Anthropic, a lab built around safety first.
• 2023: OpenAI's board fires Altman over safety concerns. He is back within days.
• 2026: Jacob Coxon resigns from Anthropic, and Amodei calls the industry's approach reckless.

Every entry on that list is a response to the same fear, and every response adds a competitor. Nobody in twelve years has answered the danger by building less. The industry's reflex when told that a technology might kill everyone is to start a company that will build it more carefully than the last one, and the last one keeps building too.

Altman said as much at the beginning, and his sentence is the honest version of the whole decade. If it cannot be stopped, the reasoning goes, better that we do it. That argument justifies founding OpenAI in 2015 and it justifies refusing to slow down in 2026, which is a sign it proves too much.

Coxon tweeted on 9 September that Anthropic and OpenAI were racing straight to self-improving superintelligence and gambling with our lives. Dozens of staff at both firms and their competitors echoed him. Days later the chief executives of the leading American labs answered a call from Amodei to slow what he termed a reckless approach, and Amodei runs one of the two companies Coxon named.

What the warnings actually bought

What the warnings have bought is regulation at a lag. Hawking spoke in 2014 and the European AI act arrived a decade later. Geoffrey Hinton resigned in 2023 and an American executive order on safety testing followed within months, though Congress produced little after it. This month Trump and Speaker Mike Johnson rejected a slowdown on the argument that a pause hands the lead to China. A week after the most-viewed warning in the field's history, no bill exists.

The same shape in another market

Readers of this site have seen this pattern in another market. Every serious dispute over how a chain should be run has ended in a fork, which is to say a second chain, not a halt. Ethereum Classic and Bitcoin Cash exist because people who disagreed about risk built their own version rather than stopping. Whether the analogy comforts you depends on what you think the downside is, and one DeepMind researcher puts it at one in three.

The money follows the same logic. We looked yesterday at how the institute supplying the most-quoted explanation of this moment was funded by a meme coin, at twenty times the size its donor intended. Warnings have become an industry with revenue, staff and lobbying budgets, and an industry rarely argues itself out of existence.

Informational material, not investment advice. The closing comparison is our reading, not a claim about anyone's intent.