The contract included the agreement, and the government is now trying to change the contract, hence the disagreement

They did consider it, got a contract that affirmed that the military would be bound by the same pre-existing terms of service as every other user, and want to resist the military's pressure to renegotiate.

Surely that might be naive but the entire issue is that they want to stick to the original contract, which is of course the purpose of a contract in the first place.

>Anthropic has repeatedly asked defense officials to agree to guardrails that would restrict its AI model... also wants to ensure Claude is not used by the Pentagon for final targeting decisions in military operations without any human involvement, one source familiar with the negotiations said. Claude is not immune from hallucinations and not reliable enough to avoid potentially lethal mistakes, like unintended escalation or mission failure without human judgment, the person said.

They explicitly allow it to be used in military operations, just not killing people without a human in the loop

source: https://www.cbsnews.com/news/pentagon-anthropic-offer-ai-unr...

The dispute is over the completely automated operation of lethal weapons powered by Anthropic's products, it has nothing to do with AI safety.

Probably already is. I'm confident that whatever us plebs get access to is less capable than what nation states do.

Lol the CEO of Palantir already bragged that occasionally his enemies have to be killed.[]

It's honestly wild watching companies with common investors, and when you dig into the details, their executives are bragging about killing their enemies. And then people argue that when surveillance is used to systematically individually stalk all of us it's magically not illegal, even though if you did that to a bunch of your ex girlfriends tracking all their movements to work and the grocery store and argued 'muh free speech to record' your ass would be in jail lickity split because there is a big difference between recording the public and stalking people while conspiring with people who are literally bragging about the killing of their enemies.

[] https://www.youtube.com/watch?v=G5gC_fParbY

"going to be"?

Always has been! Lots of math/ML/regressions already being used to make kill decisions: homing missiles, kamikaze drones, naval defense/sentries. All use a combination of computer vision, signal classification, predictive tracking. LLMs just the latest solved math problem.

Is that a good or a bad thing?

Not just people ...it will be used to kill Americans.

Military and intelligence agencies have been involved in Silicon Valley for its entire existence. They were instrumental in helping to create the first computers. The internet was a DARPA project. Facebook sprung up from a failed DARPA project called lifelog. Google received funding from the military industrial complex / intelligence agencies early in its life. These companies have always had a dual-purpose, and I highly doubt the major players in AI are any different.