Hacker News Reader

Hacker News

Comment by esperent | Hacker News Reader

Full thread

esperent

Actual article, rather than Twitter link:

https://www.nonzero.org/p/iran-and-the-immorality-of-openai

This uses this Washington Post article as a source

https://www.washingtonpost.com/technology/2026/03/04/anthrop...

(Non paywall: https://archive.is/bOJkE)

As far as I know, wasn't Claude banned from use in the Pentagon a few days ago, exactly for taking a weak stance against this kind of thing?

> Even if Amodei’s scruples had somehow magically prevented the bombing of that school, Claude would still be an accomplice to mass murder.

This point from the nonzero blog I take issue with. If they had used Google Maps to pick targets, would that make Maps an accomplice?

The people who pushed the button to launch the missiles that hit the school, and the people who ordered them to do that, are fully responsible here, not the tools they used.

View on HN

> The people who pushed the button to launch the missiles that hit the school, and the people who ordered them to do that, are fully responsible here, not the tools they used.

Absolutely. A real issue here is the normalizing of "AI scapegoating".

The real failure? Not following through on human verification of a "strong lead".

The Iran school site absolutely was _once_ a target, in the distant past - it's sited on and within a former Iranian Guard post with airstrip, etc.

The part that needed strong checking was "history since last identified as a target" - and that site has a history of disrepair and abandonment.

The debatable issue was whether the larger site did indeed store significant military assets underground, etc. which was entirely possible.

> Claude banned from use in the Pentagon a few days ago

Not exactly, you might want to reread the news to understand what's actually happening.

Didn't read the articles, but at least the planners know and understand a map.

SO... a map is static reference. A calculator is deterministic computation. An LLM is probabilistic generation

In high-stakes environments like military planning, tools that generate new claims rather than reference known data introduce a different class of risk.

Yes, everyone is responsible for their own decisions. But then circle back to risk. How can the planners be sure they aren't dealing with hallucinations, questionable data, differing outputs based on prompts, and a long list of other things...

> How can the planners be sure they aren't dealing with hallucinations, questionable data, differing outputs based on prompts, and a long list of other things...

I'm not sure they care nor do I know who holds stealth bombers accountable. We're back in the might makes right world.

> Didn't read the articles

Then kindly shut the fuck up.

Reply on news.ycombinator.com