For the record I 95% agree with you, but I'd like to try an answer at:

> I haven't seen one convincing model of AI existential risk. Can anyone offer one?

We seem to be heading down this uncomfortable trend of many people using LLM models and anthropomorphising them into properties they don't actually have, like general intelligence and trustworthiness.

So in my opinion the most likely risk here is some underpaid military man that doesn't care about the quality of their work abuses AI for it and that AI hallucinates that nukes are incoming and world war III starts. To be fair this is a fairly low risk, but it's _a_ risk.

> some underpaid military man [...] abuses AI [...] hallucinates that nukes are incoming

here it is: https://arstechnica.com/ai/2026/09/report-us-almost-boarded-...

> To be fair this is a fairly low risk, but it's _a_ risk.

it's much higher than you think, at least during Trump admin. have you seen how LLM agents are being shoved down our throats at tech companies? the government is no better in that regard, there are all incentives to skip output verification

> anthropomorphising them into properties they don't actually have, like general intelligence and trustworthiness.

Wdym? GPT-6-Astra is AGI - Jensen even said so!