On September 12, 2026, Anthropic CEO Dario Amodei published an essay urging frontier AI companies, including his own, to slow down development for fear of destroying human life as we know it in ways that we can’t even conceive of. On September 14, 2026, Anthropic posted a job listing looking for a policy design manager to handle conventional weapons, one of the ways we can very clearly conceive of how it destroys human life.
The job listing, spotted on Daybook, a jobs platform that focuses on policy and government affairs-related roles, is for a position on Anthropic’s Safeguards team, which the company says is tasked with defining and holding “the limits on how Claude can be used.” This role in particular will apparently be in charge of figuring out valid uses of Anthropic’s AI model as it relates to, ya know, bombs and such, and what crosses the line. The fact that there is a negotiation would suggest that there are acceptable ways to use Claude to blow stuff up, which does not feel like it jives with a company seemingly concerned with the existential threats of AI.
“A crisp boundary between acceptable and harmful requests in this domain is difficult, because the underlying components are dual-use: the same capabilities that serve civilian engineering and research can also contribute to a weapons system,” the job listing states. “Conventional weapons are increasingly defined by software, and the risks reach well beyond firearms: from a model operating a weapons system or writing guidance code, to the weaponization of dual-use platforms. The role spans every weapon class, up to autonomous systems that select and engage targets without human authorization.”
Anthropic has, to its credit, walked the walk when it comes to its red line on autonomous weapons systems. It blew up its relationship with the Pentagon by refusing to allow the Department of Defense to use its models for weapons that wouldn’t require human intervention.
But the company has been much squishier on weapons that still require a person to hit the deploy button. Per Reuters, Anthropic’s AI tools were used in the United States’ attack on Iran. The technology was also allegedly used in America’s raid on Venezuela that resulted in the capture of Venezuelan President Nicolás Maduro, according to the Wall Street Journal.
Amodei’s alarming letter doesn’t refer to weapons systems, autonomous or otherwise—which are uses in which AI is already being integrated and killing people. Instead, it focuses on theoretical threats that have yet to manifest, like hacks that take down the entire internet or the development of a biological weapon. If you’re worried that AI could kill everyone, you might expect your policy on conventional weapons to be a hard no—not a line drawn by managers that lets you claim responsible development without jeopardizing lucrative defense contracts.
Gizmodo reached out to Anthropic for comment on how its position on conventional weapons fits into Amodei’s call for AI safety, but did not receive a response at the time of publication.
Read the full article here
