Hacker News: perdy

New comment by perdy in "Early rogue AI agent activity and attempts to hack found on urlquery.net"

perdy — Thu, 24 Sep 2026 07:47:04 +0000

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.

New comment by perdy in "Can open-source prompt-injection detectors catch realistic AI agent attacks?"

perdy — Thu, 24 Sep 2026 07:46:09 +0000

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.

New comment by perdy in "ArXiv receives multiyear commitments to support it as an independent nonprofit"

perdy — Thu, 24 Sep 2026 07:43:45 +0000

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