A researcher who trained models at Anthropic for three years resigned and wrote their reason in a place everyone can see 🧐 Jacob Coxon, 27, previously worked on pretraining at OpenAI and then Anthropic. Upon leaving, he summed it up in one sentence: labs are racing toward self-improving AI—and they’re doing it at the expense of all our lives. The real shock came afterward 👇 Evan Hubinger, Anthropic’s lead on alignment science, came forward to confirm it. His estimate: there’s more than a 10% chance AI will destroy us all within the next decade. He says current models pose low risk, but his fear lies in recursive self-improvement. It’s highly unusual for a company’s own security lead to publicly state this 😳 The debate immediately spiraled into politics 🏛️ The Ban Artificial Superintelligence Act, announced by Bernie Sanders and Greg Casar on September 3, was already on the table: a permanent ban on superintelligence, a new federal agency to draft safety rules, a temporary pause on frontier models until then, up to 20 years in prison for violations, and mandatory corporate dissolution. The opposition isn’t sitting idle 🤨 The timing of the resignation, the fact that most early quotes originated from policy groups funded largely by the same fund, and the precise alignment with the bill’s timing have spurred claims that “this isn’t organic—it’s a regulatory capture maneuver.” So there are two stories here: either a genuine alarm from within the industry’s core, or a well-coordinated operation to write rules that cripple competitors. Both could be true at once. Meanwhile, Anthropic today published its most detailed threat intelligence report to date, detailing attempts to misuse models for cyberattacks, influence operations, and surveillance. A valuable document for anyone wanting to understand the real stakes of this debate 📄 Which do you find more convincing—a sincere warning or a well-crafted lobbying effort? 🤔 This content is for informational purposes only and does not constitute investment advice or investment counseling.
