A rumor, not a religion: 'Anthropic engineers worship Claude' says more about the IPO window than the lab

The claim is striking and the sourcing is thin. Futurism reported on September 17 that some Anthropic engineers are said to "worship" Claude, relaying a rumor that, per The Spectator's Sean Thomas, is "running rampant in Silicon Valley." Thomas frames it as an old science-fiction prophecy coming true: "If we build a superintelligence, we will first fall in love with it. Then we will worship it. Finally we will make sacrifices to it." What the piece does not contain is a named engineer, a document, an incident or a witness. The load-bearing word in the headline is "rumored," and it should stay there — we are reporting that a claim is circulating, not that it is true.

The harder material in the same article is duller and, we'd argue, more informative. Per The Information, getting hired at Anthropic requires prospective employees to pledge allegiance to AI safety and sit a psychological profiling test. Per the BBC last month, some staff work up to 90-hour weeks inside what it described as a "grueling" culture with "cut-throat" performance reviews and frequent layoffs. And Futurism notes the commercial backdrop: both OpenAI and Anthropic are aiming to go public before the end of the year. None of that is mystical. All of it explains the story better than theology does.

There is a market for divinity, and it has two sides. Sellers benefit: a company conjuring something god-like is easier to value than one shipping a very good text-and-code tool, which is why Sam Altman was calling it a "magic intelligence in the sky" back in 2023 and why Mark Zuckerberg found it worth mocking rivals in 2024 for acting "as if they kind of think they're creating God." Critics benefit too: a lab where engineers allegedly kneel is a far better read than one where they argue about eval harnesses. Futurism, to its credit, flags the incentive itself — there is a lot to gain from convincing investors you have summoned a mightier spiritual force than the competition, "even if reality is far more mundane than that."

This is where we'd push the analysis further than the piece does. The real damage of deification is epistemic, not spiritual. A god cannot be benchmarked. Worship — or its mirror image, apocalyptic dread — moves a model out of the domain where it can be measured on coding benchmarks, agentic tasks and unguided cybersecurity evals, and into a domain where no claim about it can be falsified. That is a gift to marketing departments and a disaster for safety work, which is nothing but auditing, red-teaming and unglamorous measurement. Every assertion that a system is divine is, functionally, an assertion that measurement is beside the point.

Two threads here deserve to be taken seriously even though the cult rumor doesn't. The first is user harm: the article's account of AI psychosis — people spiraling into the belief that they are prophets, or a divine figure themselves — describes documented damage to ordinary users, not gossip about engineers. The second is labor. When a workplace asks for mission-level conviction and also runs 90-hour weeks with cut-throat reviews, sincere belief and career self-preservation become impossible to distinguish from the outside. That is a management story, and it is the one worth reporting out with named sources.

Our reading: treat this as a reminder, not a revelation. Nothing here changes the long horizon we hold to — these systems are on a credible path to helping erase diseases, extend healthy life and generate enough abundance that people can work on what they actually care about. But that road runs through replication, evaluation and boring engineering, not through revelation. The correct posture toward a very capable model is the same as toward a very capable colleague: use it, check it, and be willing to tell it that it's wrong. Neither kneel nor panic. A tool you worship is a tool you stop auditing — and auditing is exactly what gets us to the good ending.