10% Chance AI Will Kill Humanity Within a Decade - Anthropic Employees are Evil AND Stupid

A 28-minute monologue on Eli the Computer Guy’s own show, reacting to the week’s headline: an Anthropic alignment researcher putting the chance that AI kills all humans at more than 10% within a decade. His position has two halves. AI will not exterminate humanity, and anyone who believes it will and still goes to work on it is, in his word, worthless.

The technical half rests on one idea: intelligence is a single component of the system needed to do anything, whether that is replacing workers, curing disease or wiping out the species. Every outcome needs processes and mechanisms around it, and a data centre has none of its own. He grants that automated systems could cause mass unintended deaths, or start a war by setting each other off, but not a deliberate extinction. He blames the belief on insulation: a culture taught that the world is built with code, which forgets the plumbers, farmers and builders who actually built it. His test is a comparison. The world changed between 1995 and 2001 with Windows 95, and between 2008 and 2014 with the iPhone. Take away the people selling AI and he doubts most people would notice what it changed between 2020 and 2026. Outside programming, he says, it has been “a bit of a damp fart”.

The moral half takes the researchers at their word. If you think your work might kill everyone and you keep doing it for salary and equity, he says, that says worse things about you than about the technology. He compares it to chemists learning that DDT was thinning birds’ eggshells and responding by making it twice as strong. He argues that the only consistent act for a true believer would be sabotage, then says plainly that he is urging no one to do it, because the premise is stupid. Much of the episode is invective in this vein, aimed at lab staff and at Dario Amodei.

What he quotes checks out. Jacob Coxon announced on Tuesday 8 September 2026 that he had resigned from Anthropic, saying neither Anthropic nor OpenAI is acting responsibly and that the labs are “gambling with our lives”. Evan Hubinger replied on X that he puts the chance above 10% within the next decade and that Anthropic does not yet have a plan to solve alignment for superintelligence, and later said his worry is superintelligence arising from recursive self-improvement rather than today’s models. CNBC carried the same quotes. The line about systems “fully building their own successors” is from Anthropic’s post of 4 June 2026, “When AI Builds Itself”. Two of his details slip. The iPhone launched in June 2007, a year before his window opens. And the scene of a war set off by a radar mistaking a condor for a missile, which he credits to James Morrow’s Bible Stories for Adults, could not be found in that collection.