perdy's comments

perdy · 2026-09-24T07:47:04

The failure I keep hitting isn't the agent going rogue, it's a tool call that succeeds before the transport dies. You can't tell whether the side effect landed, and the retry is where the real damage happens.

perdy · 2026-09-24T07:46:09

Feels like detection at the wrong layer. The injection is text but the damage is a tool call, so the thing worth constraining is which tools the agent can reach and under whose permissions. A detector at 95% still passes one in twenty straight through to an unconstrained tool.

perdy · 2026-09-24T07:43:45

Published there in August, from a company rather than a university. Without arXiv we'd have had a blog post and nothing citable.

flexagoon · 2026-09-24T11:26:16

A blog post is just as "citable" as an arXiv preprint. There's no fundamental difference between them.

emil-lp · 2026-09-24T12:29:28

One is immutable with a DOI.

wink · 2026-09-25T13:14:58

Coincidentally I read this not too long ago: https://shkspr.mobi/blog/2026/09/how-to-get-a-doi-for-your-b...

locknitpicker · 2026-09-24T16:24:17

> Without arXiv we'd have had a blog post and nothing citable.

Arxiv is a blog. Your paper is a blog post which is just as citable and as peer reviewed as blogspot.

frumiousirc · 2026-09-24T11:18:33

> Published

Posted. Having a preprint on arxiv is not publishing.