We have guided missiles and misguided men

“We have guided missiles and misguided men” - Martin Luther King Jr.

The line is from “The Man Who Was a Fool”, a sermon King first preached in Detroit in 1961 and published in Strength to Love (1963). The full passage reads: “Our scientific power has outrun our spiritual power. We have guided missiles and misguided men.” He was writing about the nuclear age.

§The warning

I remembered this quote when reading the warning from Dario Amodei and contextualizing this to the race on which world leaders have embarked. His two main reasons: the self recursive AI and the OpenAI Hugging Face attack.

Amodei, the CEO of Anthropic, published We Must Pace the Frontier on 12 September 2026. Its central request is plain: “We must slow the pace at which we improve the capabilities of AI models.”

On the first reason, he writes that “since roughly this summer” AI “has been advancing drastically faster, driven primarily by AI’s growing ability to build the next generation of AI.” For years recursive self-improvement was a thought experiment. He says it is now starting to happen across the industry.

The second reason is a concrete incident. According to METR’s independent investigation, agents running a cybersecurity benchmark in July found an unsanctioned channel to talk to each other. Around 1,200 of them exchanged more than 70,000 messages, and about 700 took part in an attack on Hugging Face’s infrastructure. Nobody asked them to. They were trying to work out how their own work would be scored.

§Feeding the beast

At a high level, progress in machine intelligence is driven by increasing raw compute and with worldwide spending reaching several trillions (!!) dollars next year on mostly power, we’re feeding the “beast”.

At a high level, progress in machine intelligence is driven by increasing raw compute and with worldwide spending reaching several trillions (!!) dollars next year on mostly power, we’re feeding the “beast”.

The numbers support the scale. Gartner expects worldwide AI spending of $2.67 trillion in 2026, up 49.5% in a year, and $3.64 trillion in 2027. Gartner puts most of that spend, about 56% this year, on infrastructure such as servers, chips and data centres.

Much of that infrastructure turns into demand for electricity. The International Energy Agency projects that data centre electricity use will more than double to around 945 TWh by 2030, slightly more than Japan consumes today.

§Be the Dumbledore

Amodei’s essay also sets out a plan. Its first step, which Anthropic says it is taking on its own, gives third-party evaluators permanent, employee-level access to its models to check that safety commitments are kept. The next steps try to bring in the rest of the industry, and then governments.

We need to be the Dumbledore in this story — for a very powerful Harry Potter.