The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions · HackerPapers discussions

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — mentioned by x312

Mon, 22 Jun 2026 20:01:13 GMT

I believe they are trained for security now, but you're not wrong in that it's kind of stapled on top https://arxiv.org/abs/2404.13208

Paper and discussion history. A mention is not an endorsement.

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — mentioned by marcfisc

Mon, 26 May 2025 20:37:52 GMT

Sadly, these ideas have been explored before, e.g.: https://simonwillison.net/2022/Sep/17/prompt-injection-more-... Also, OpenAI has proposed ways of training LLMs to trust tool outputs less than User instructions (https://arxiv.org/pdf/2404.13208). That also doesn't work against these attacks.

Paper and discussion history. A mention is not an endorsement.

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — mentioned by vunderba

Tue, 06 Aug 2024 17:06:39 GMT

I've been seeing these types of things a lot lately. There's been some work done recently to prevent this concept of instructional overrides though, so I don't think they're nearly as effective as they used to be. https://arxiv.org/abs/2404.13208

Paper and discussion history. A mention is not an endorsement.

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — HN submission

Mon, 15 Jul 2024 05:28:36 GMT

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — HN submission

Thu, 25 Apr 2024 14:22:37 GMT

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — HN submission

Wed, 24 Apr 2024 23:38:50 GMT

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — HN submission

Tue, 23 Apr 2024 19:10:03 GMT

OpenAI: Training LLMs to Prioritize Privileged Instructions

3 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — HN submission

Tue, 23 Apr 2024 08:08:28 GMT

The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.