2026-07-28 03:35 UTC
@muddle@infosec.exchange @vrek@mastodon.social @lzg@mastodon.social I just did. 'Guardrails' are no more than direct instructions to the LLM, in natural language, included as part of the system prompt given before any user instruction, eg: 'Please do this. Please do not do that'. It's literally that stupid. And also that fragile. Hence the ease with which every LLM can be caused to misbehave, b/c suitably crafted user input can always override instructions in the system prompt.
Replies (1)
-
@muddle@infosec.exchange 2026-07-28 03:38
@conniptions@mastodon.social @vrek@mastodon.social @lzg@mastodon.social So I think we got back to the original point, which is that guardrails aren't/don't. Fundamentally, when LLM-bros talk about "guardrails" it's a form of epistemic assault. They don't know what the word means, and even they, if they think about it for just a second should be able to figure out that they're a phantom. Not real.