Conspiracy
Anthropic CEO Warns AI Will Kill Everyone Then Asks Congress Not To Regulate Anthropic
By Tah Foil, . BSN Network. Satire.
On September 12, Dario Amodei, the chief executive officer of Anthropic, published an essay titled "The Gentle Singularity," in which he described a future in which artificial intelligence either elevates the human species to near-immortality or destroys it entirely, and then recommended, at considerable length, that the entity best positioned to prevent the second outcome was Anthropic.
The essay runs approximately 15,000 words. Anthropic has a current valuation of approximately $61 billion.
The publication came one week after Jack Clark, a co-founder of Anthropic who left the company in 2021, posted a thread on X stating that he believed AI had a realistic probability of killing all humans within a decade. Clark phrased this carefully, methodically, and with citations, which is the kind of thing that tends to go very viral and also tends to generate a Senate hearing with a title like "AI: What Is It Doing."
Lawmakers responded with the urgency appropriate to a species-level threat, which is to say they responded the way they respond to everything: by scheduling meetings, floating the idea of a subcommittee, and asking their staff to look into whether any existing legislation could be lightly modified to cover the robot apocalypse without requiring anyone to write new legislation.
Senator Maria Cantwell told reporters that Congress is "taking this very seriously" and would be "exploring a range of frameworks." She did not specify which frameworks. She did not specify a timeline. She did specify that she had read "significant portions" of the Amodei essay, which her office later clarified meant the introduction and a summary she received by email.
Amodei's essay argues that AI could, within the next three to five years, compress a century of medical progress into a single decade, cure most diseases, and generate enough economic growth to end global poverty. It also argues that if any of this is done incorrectly, everyone dies. The essay does not dwell on the tension between these two positions, preferring instead to treat them as a project management challenge.
Crucially, the essay recommends that regulation focus on "frontier AI safety," a term Anthropic coined, measured by benchmarks Anthropic helped design, assessed by a proposed federal body that Anthropic executives have already offered to help staff.
Marcus Delray, a former DARPA contractor who now consults for three of the companies he is being asked to evaluate, told this reporter that the framework was "genuinely rigorous" and that he saw no conflict of interest worth reporting. He then said the same sentence again, slightly louder.
Meanwhile, in a separate thread, Clark clarified that his estimate of AI killing everyone was not a prediction but a "probability distribution," which is the kind of phrase that makes a claim sound more scientific while also making it impossible to be wrong about. He received 4.2 million impressions. Congress received the thread as a printed PDF, stapled in the wrong order.
The House Science Committee announced a hearing for November, in which it will invite several AI company executives to explain, in terms a congressman can understand, what a large language model is. This will be the seventh such hearing since 2022. At the conclusion of the sixth hearing, Senator Ron Johnson asked an OpenAI representative whether the AI could "be taught to respect the Constitution," and the representative said yes, because the alternative answer ends careers.
A working group was formed. The working group has fourteen members. Four of them are the same person across different committees, which is not possible but is what the paperwork reflects.
Clark, reached for comment through a representative, said he stood by his original thread and was "cautiously hopeful" that government action would follow. He did not specify what action. He did not specify cautiously hopeful compared to what.
The essay remains available on Anthropic's website, next to a button inviting you to start a free trial of Claude.
The story we are making fun of: https://politifact.com/article/2026/sep/14/congress-artificial-intelligence-bills-anthropic/