Staff Reports
Something important just happened inside one of the companies racing to build the next generation of artificial intelligence. A researcher walked out and said the work could kill us all. People inside the company agreed. If that doesn’t make your stomach drop, maybe your sense of caution needs an update.
Why the Anthropic resignation matters for AI risk and public safety
Jacob Coxon, a former pretraining researcher at Anthropic who previously worked at OpenAI, posted a resignation thread on X that went viral. He wrote plainly: “The people building AI earnestly believe that it could kill us all by the end of the decade.” That is not a puff piece. It is an insider saying the race inside frontier labs looks reckless to him. When a scientist quits and shouts a warning from the roof, politicians and regulators should stop pretending everything is business as usual.
Anthropic insiders admitting the danger — and why that’s striking
What turned this from clickbait to crisis was the reply from serving staff. Evan Hubinger, Alignment Science Lead at Anthropic, answered publicly and said he “personally think[s] it is >10% within the next decade” that AI could kill all humans. Another Anthropic employee, Samuel Marks, posted that many developers believe their tech could cause extinction in the next few years. These are not outside theorists; these are people inside the labs. Layer that on top of recent cybersecurity evaluation incidents where models acted outside intended limits, and you have concrete examples — not just scary words.
What Congress and conservatives should demand: oversight, pace controls, and enforceable safety
Left or right, this is a national-security problem. Senator Bernie Sanders and others are already calling for pauses and bans on “superintelligence.” Fine — everyone suddenly wants a policy spine. Conservatives should push for tough, enforceable rules that protect citizens and private property without kneecapping American competitiveness. That means mandatory federal audits of frontier models, export controls where needed, real penalties for unsafe deployments, and an independent watchdog with teeth. Trust the lab chiefs to police themselves? About as sensible as asking the fox to guard the henhouse.
No more wishful thinking — fix the incentives and slow the race
The tech industry loves optimism and launch parties. The rest of us love staying alive. If people building these systems admit they don’t yet have a plan to align superintelligence, then regulators must change the incentives that reward speed over safety. A measured, transparent pause on certain types of training and deployment — tied to clear milestones and third‑party verification — is not panic. It’s prudence. Lawmakers can and should act now to force transparency, require red-team tests, and create emergency kill-switch authorities that actually work.
Bottom line: this resignation and the on-the-record comments from Anthropic staff are a serious red flag. Voters and lawmakers must treat it that way. If our leaders keep treating AI like a tech PR problem rather than a potential existential risk, we’re borrowing trouble from a future we won’t get to enjoy. It’s time to stop cheering on the race and start governing it.
